User Accuracy Research Articles

Geodl: An R package for geospatial deep learning semantic segmentation using torch and terra.

Convolutional neural network (CNN)-based deep learning (DL) methods have transformed the analysis of geospatial, Earth observation, and geophysical data due to their ability to model spatial context information at multiple scales. Such methods are especially applicable to pixel-level classification or semantic segmentation tasks. A variety of R packages have been developed for processing and analyzing geospatial data. However, there are currently no packages available for implementing geospatial DL in the R language and data science environment. This paper introduces the geodl R package, which supports pixel-level classification applied to a wide range of geospatial or Earth science data that can be represented as multidimensional arrays where each channel or band holds a predictor variable. geodl is built on the torch package, which supports the implementation of DL using the R and C++ languages without the need for installing a Python/PyTorch environment. This greatly simplifies the software environment needed to implement DL in R. Using geodl, geospatial raster-based data with varying numbers of bands, spatial resolutions, and coordinate reference systems are read and processed using the terra package, which makes use of C++ and allows for processing raster grids that are too large to fit into memory. Training loops are implemented with the luz package. The geodl package provides utility functions for creating raster masks or labels from vector-based geospatial data and image chips and associated masks from larger files and extents. It also defines a torch dataset subclass for geospatial data for use with torch dataloaders. UNet-based models are provided with a variety of optional ancillary modules or modifications. Common assessment metrics (i.e., overall accuracy, class-level recalls or producer's accuracies, class-level precisions or user's accuracies, and class-level F1-scores) are implemented along with a modified version of the unified focal loss framework, which allows for defining a variety of loss metrics using one consistent implementation and set of hyperparameters. Users can assess models using standard geospatial and remote sensing metrics and methods and use trained models to predict to large spatial extents. This paper introduces the geodl workflow, design philosophy, and goals for future development.

Mastering Gesture-Based Screen Readers on Mobile Devices - Exploring Teaching and Practice Strategies.

Gesture-based screen readers like VoiceOver or TalkBack provide visually impaired users with a means to interact with digital content. However, there is a significant lack of both strategies and resources for teaching the use of these screen readers, and standardized teaching guidelines are notably absent. Furthermore, there is no free, universally designed, and accessible app for practicing gestures in mobile screen readers. This study aims to identify best practice strategies for teaching and practicing the use of gesture-based screen readers among visually impaired users, based on observations from an IT course directed at visually impaired individuals. Moreover, we present common challenges related to usability, attitudes, emotions, technical aspects, and user guidance and education, as well as key traits and facilitators for learning gesture-based screen readers. Lastly, we assess the feasibility of an app to practice gestures and propose a framework for a gesture practice app to enhance user accuracy and patience.

The illusion of success: Test set disproportion causes inflated accuracy in remote sensing mapping research

In remote sensing mapping studies, selecting an appropriate test set to accurately evaluate the results is critical. An imprecise accuracy assessment can be misleading and fail to validate the applicability of mapping products. Commencing with the WHU-Hi-HanChuan dataset, this paper revealed the impact of sample size ratios in test sets on accuracy metrics by generating a series of test sets with varying ratios of positive and negative sample size to evaluate the same map. A rigorous approach for accuracy assessment was suggested, and an example of tea plantations mapping is used to demonstrate the process and analyse potential issues in traditional approaches. A scale factor (λ) was constructed to measure the discrepancy in sample size ratios between test sets and actual conditions. Accuracy adjustment formulas were developed and applied to adjust the accuracy of 42 previous maps based on the λ. Results showed a higher ratio of positive to negative sample size in test set led to inflated user’s accuracy (UA), F1-score (F1) and overall accuracy (OA), but had little impact on producer’s accuracy. When the ratio aligned with that in the target area, the UA, F1, and OA closely matched the true values, indicating the proportion of positive and negative samples in test set should be consistent with that in actual situation. The accuracies reported by the traditional approaches including test set sampling from labelled data and 5-fold cross validation were far from the true accuracy and could not reflect the performance of the map. Among 42 previous maps, nearly 60% of the maps had UAs overestimated by 10%, and 9.5% of the maps had UAs and F1s deviations of more than 25%. The conclusions of this study provide a clear caution for future mapping research and assist in producing and identifying truly excellent maps.

Mapping Polylepis Forest Using Sentinel, PlanetScope Images, and Topographical Features with Machine Learning

Globally, there is a significant trend in the loss of native forests, including those of the Polylepis genus, which are essential for soil conservation across the Andes Mountain range. These forests play a critical role in regulating water flow, promoting soil regeneration, and retaining essential nutrients and sediments, thereby contributing to the soil conservation of the region. In Ecuador, these forests are often fragmented and isolated in areas of high cloud cover, making it difficult to use remote sensing and spectral vegetation indices to detect this forest species. This study developed twelve scenarios using medium- and high-resolution satellite data, integrating datasets such as Sentinel-2 and PlanetScope (optical), Sentinel-1 (radar), and the Sigtierras project topographic data. The scenarios were categorized into two groups: SC1–SC6, combining 5 m resolution data, and SC7–SC12, combining 10 m resolution data. Additionally, each scenario was tested with two target types: multiclass (distinguishing Polylepis stands, native forest, Pine, Shrub vegetation, and other classes) and binary (distinguishing Polylepis from non-Polylepis). The Recursive Feature Elimination technique was employed to identify the most effective variables for each scenario. This process reduced the number of variables by selecting those with high importance according to a Random Forest model, using accuracy and Kappa values as criteria. Finally, the scenario that presented the highest reliability was SC10 (Sentinel-2 and Topography) with a pixel size of 10 m in a multiclass target, achieving an accuracy of 0.91 and a Kappa coefficient of 0.80. For the Polylepis class, the User Accuracy and Producer Accuracy were 0.90 and 0.89, respectively. The findings confirm that, despite the limited area of the Polylepis stands, integrating topographic and spectral variables at a 10 m pixel resolution improves detection accuracy.

The first 10-m China’s national-scale sandy beach map in 2022 derived from Sentinel-2 imagery

ABSTRACT Sandy beaches are at the frontline of resisting continuous sea level rise associated with anthropogenic climate change. However, accurate and comprehensive spatial information for monitoring, utilizing, and protecting sandy beaches is still lacking at the national or above scales. This study, for the first time, addresses this gap by collecting cloud-free, low-tide Sentinel-2 images in 2022 to map 10-m sandy beaches across China using the image classification method. We adopted the Support Vector Machine to derive the spatial distribution of sandy beaches, assess accuracy, and analyze spatial characteristics. Our results demonstrate the efficiency of the SVM model in mapping sandy beaches (User's accuracy: 96%, Kappa coefficient: 0.93). We identified 3,444 beaches in China, with a total length of 3,187.57 km, an average width of 69.93 meters, and a total area of 217.43 km², constituting 24.16% of the national coastline. Notably, Guangdong, Taiwan, and Hainan provinces are rich in beach resources, whereas Macao, Shanghai, Tianjin, and Jiangsu provinces have relatively fewer beach resources. Further, our results outperform the existing OpenStreetMap beach dataset. Our developed 10-m beach database is crucial for analyzing potential beach risks, uncovering socioeconomic values of beach resources, and promoting the sustainable coastal zone development in China.

Global mapping of oil palm planting year from 1990 to 2021

Abstract. Oil palm is a controversial crop, primarily because it is associated with negative environmental impacts such as tropical deforestation. Mapping the crop and its characteristics, such as age, is crucial for informing public and policy discussions regarding these impacts. Oil palm has received substantial mapping efforts, but accurate and up-to-date oil palm maps of both extent and age are essential for monitoring impacts and informing concomitant debate. Here, we present a 10 m resolution global map of industrial and smallholder oil palm, developed using Sentinel-1 data for the years 2016–2021 and a deep learning model based on convolutional neural networks. In addition, we used Landsat-5, Landsat-7, and Landsat-8 to estimate the planting year from 1990 to 2021 at a 30 m spatial resolution. The planting year indicates the year of establishment of the current (as of 2021) oil palm plantation by means of either newly planted or replanted oil palm in an existing oil palm plantation. We validated the oil palm extent layer using 18 812 randomly distributed reference points. The accuracy of the planting-year layer was assessed using field data collected from 5831 industrial parcels and 1012 smallholder plantations distributed throughout the global oil palm growing area. We found oil palm plantations covering a total mapped area of 23.98 Mha, and our area estimates are 16.82 ± 0.19 Mha of industrial oil palm and 7.37 ± 0.25 Mha of smallholder oil palm worldwide. The producer's and user's accuracy are 91.0 ± 2.5 % and 91.8 ± 1.2 % for industrial plantations and 71.4 ± 0.7 % and 72.4 ± 1.8 % for smallholders; these values represent an improvement compared to a previous global oil palm dataset, particularly in terms of omission of oil palm. The overall mean error between the estimated planting year and the field data was −0.24 years, and the root-mean-square error was 2.65 years, but the agreement was lower for smallholders. Mapping the extent and planting year of smallholder plantations remains challenging, particularly for wild and sparsely planted oil palm, and future mapping efforts should focus on these specific types of plantations. The average oil palm plantation age was 14.1 years, and the area of oil palm over 20 years old was 6.28 Mha. Given that oil palm plantations are typically replanted after 25 years, our findings indicate that this area will require replanting within the coming decade, starting from 2021. Our dataset provides valuable input for optimal land use planning to meet the growing global demand for vegetable oils. The global oil palm extent layer for the year 2021 and the planting-year layer from 1990 to 2021 can be found at https://doi.org/10.5281/zenodo.13379129 (Descals, 2024a).

Optimum Combination of Spectral Variables for Crop Mapping in Heterogeneous Landscapes based on Sentinel-2 Time Series and Machine Learning

Abstract. This article aimed to determine a workflow for more efficient large-scale crop mapping using a time series of images from the Sentinel-2 Satellite, statistical methods of attribute selection, and machine learning. The proposed methodology explores the best possible combination of spectral variables related to vegetation (16 vegetation indices in the RGB, NIR, SWIR, and Red Edge regions) to characterize different spectro-temporal profiles of Land Use and Land Cover (LULC) in spatially heterogeneous landscapes. First, we applied a data dimensionality reduction analysis using the PCA (Principal Component Analysis) method. Subsequently, the variables that showed the highest statistical correlation between each other were used in the spectro-temporal classification process, using the Random Forest, TempCNN, and LightTAE algorithms, following three different strategies: C1 (ALL), C2 (BE + IV (Red Edge)) and C3 (BE + IV (without Red Edge)), where ALL – All variables; BE – Spectral Bands; IV – Vegetation Indices. Given the results found, the C2 classification scenario (with bands B3, B4, B5, B6, B7, B8, and B8A and the NDRE1, RESI, and MSR indexes) demonstrated the best LULC classification accuracy at the crop pattern level, compared to the other scenarios, with average values of 0.91, 0.88, 0.91, 0.89, and 0.89 (Global Accuracy, Producer Accuracy, User Accuracy, Kappa index, and F1-Score, respectively, for the TempCNN model), the which emphasized the importance of both qualitative and quantitative variability of sampling data and variables based on the Red Edge region for improving LULC classification processes in large-scale heterogeneous landscapes.

Mapping aquaculture in inland continental areas of Brazil using machine learning on the Google Earth Engine

Aquaculture has played a significant and growing role in the provision of food, food security, and job creation, while simultaneously establishing itself as an important agricultural activity subject to increasingly stringent market practices. The generation of spatial information on aquaculture activity in Brazil can support more assertive planning of new enterprises as well as enable the monitoring of this activity over time, fulfilling an agenda of sustainable aquaculture growth in the country. The aim of this article was to map aquaculture areas in the state of Paraná (Brazil) by developing an automated methodology for the extraction of aquaculture ponds. The methodology was implemented on the Google Earth Engine (GEE) platform and utilized satellite images with a spatial resolution of 4.77 m from Norway’s International Climate and Forest Initiative (NICFI) Program – Planet. The process involved pixel-oriented classification of satellite images and the utilization of the Random Forest (RF) algorithm for classification, based on 1200 training samples. The accuracy of the mapping and the algorithm was evaluated using Producer Accuracy (PA), User Accuracy (UA), and Overall Accuracy (OA). The mapping achieved an OA of 0.87, with a PA of 0.89 and a UA of 0.96 for the aquaculture class, and a PA of 0.66 and a UA of 0.41 for the non-aquaculture class. 42,369 aquaculture ponds were identified in the study area, covering a total area of 11,515 ha. The ponds had sizes smaller than 0.7 ha and non-circular shapes (compactness 0.70). Around 40% of the identified aquaculture tanks were concentrated in just two out of the ten analyzed mesoregions, accounting for 40% of the water surface area used for aquaculture in Paraná. Considering that the results obtained show potential to fill the gap in geospatial information about state aquaculture, our methodology would provide a broad understanding of the state’s aquaculture framework, definition of regions with capacity for the expansion of aquaculture ponds, accuracy in estimating the state’s productive space, monitoring the improper expansion of aquaculture areas, as well as establishing locations where the pressure on water resources by aquaculture prevents the expansion of this activity. Therefore, this study could benefit public and private organizations, and other stakeholders, by systematically monitoring aquaculture production in the country’s leading fish farming state, contributing to assertive decisions in regions where aquaculture production needs to expand sustainably.

Comparative use of UAV and Satellite Images in discrimination and estimation of cashew plantation areas

Cashew plantations generate significant interest in Benin due to their high socioeconomic value for the population. A thorough understanding of the spatial distribution of these plantations is crucial for comprehending their environmental and socioeconomic impacts. In this study, various types of multi-sensor imagery were compared to assess each sensor's capabilities in mapping plantation areas. The study was conducted in the Savè commune, a major industrial cashew-producing region. Multispectral sensors from Landsat-8 Operational Land Imager (OLI), Sentinel-2A, and UAV multispectral platforms, along with ground surveys, were fused and classified using the Random Forest algorithm. The study results allowed for the assessment of uncertainties associated with different platforms in detecting cashew plantations in the test area. Classification using Random Forest algorithms on UAV, Sentinel, and Landsat platform images yielded overall accuracies of 83%, 65%, and 48%, respectively. Producer and user accuracies were 94% and 75% for the UAV platform, 98% and 71% for the Sentinel platform, and 91% and 77% for the Landsat platform in cashew tree detection. This study demonstrates the complementarity among various platforms in detecting and mapping cashew plantations.

Indicator element selection and lithological mapping using deep learning methods in the Dahongliutan area, NW China

Rare metal resources are extensively used in the emerging energy field, making the security and sustainability of rare metal supply chains critical issues. Pegmatite-type rare metal deposits are a significant source of rare metal resources. Geochemistry is one of the most direct and effective methods of mineral exploration. In this study, whole-rock geochemical data from the Akesayi region, located in the Western Kunlun area of China, were used to identify the indicative elements of pegmatite automatically. Based on the stream sediment geochemical data, various deep learning models have been employed to achieve automatic lithological identification of the area. The results indicate that a novel interpretable model using SHapley Additive exPlanations (SHAP) and eXtreme Gradient Boosting (XGBoost) was employed to select indicative elements for the pegmatite in the Akesayi region, identifying Ta and Rb as key elements. The state-of-the-art application of deep-learning algorithms for lithological mapping has proven to be highly effective. Among the four approaches, the ensemble strategy integrating 1D convolutional neural networks, 2D3D convolutional neural networks, and dual-branch neural networks yields the best lithological mapping results. This approach resulted in a total classification accuracy of 90.422 %, an average accuracy of 90.502 %, a Kappa coefficient of 89.643 %, and a user accuracy of 65.530 % for the pegmatite lithological unit. These results demonstrate that the proposed model can provide robust technical support for the exploration of rare metal pegmatites in regions with challenging natural conditions and limited research.

SSAT: Active Authorization Control and User’s Fingerprint Tracking Framework for DNN IP Protection

As training a high-performance deep neural network (DNN) model requires a large amount of data, powerful computing resources and expert knowledge, protecting well-trained DNN models from intellectual property (IP) infringement has raised serious concerns in recent years. Most existing methods using DNN watermarks to verify the ownership of the models after IP infringement occurs, which is reactive in the sense that they cannot prevent unauthorized users from using the model in the first place. Different from these methods, in this article, we propose an active authorization control and user’s fingerprint tracking method for the IP protection of DNN models by utilizing sample-specific backdoor attack. The proposed method inversely and multiplely exploits sample-specific trigger as the key to implement authorization control for DNN model, in which the generated triggers are imperceptible and sample-specific for clean images. Specifically, a U-Net model is used to generate backdoor instances. Then, the target model is trained on the clean images and backdoor instances, which are inversely labeled as wrong classes and correct classes, respectively. Only authorized users can use the target model normally by pre-processing the clean images through the U-Net model. Moreover, the images processed by the U-Net model will contain unique fingerprint that can be extracted to verify and track the corresponding user’s identity. This article is the first work that utilizes the sample-specific backdoor attack to implement active authorization control and user’s fingerprint management for DNN model under black-box scenarios. Extensive experimental results on ImageNet dataset and YouTube Aligned Face dataset demonstrate that the proposed method is effective in protecting the DNN model from unauthorized usage. Specifically, the protected model has a low inference accuracy (1.00%) for unauthorized users, while maintaining a normal inference accuracy (97.67%) for authorized users. Besides, the proposed method can achieve 100% fingerprint tracking success rates on both the ImageNet and YouTube Aligned Face datasets. Moreover, it is demonstrated that the proposed method is robust against fine-tuning attack, pruning attack, pruning attack with retraining, reverse-engineering attack, adaptive attack, and JPEG compression attack. The code is available at https://github.com/nuaaaisec/SSAT .

OPEN-PIT LIMESTONE MINING AREAS MAPPING AND ASSESSMENT USING RANDOM FOREST ALGORITHM FOR SUSTAINABLE DEVELOPMENT

Limestone excavation is prone to land use/cover change, which impacts the ecosystem. The objective of this paper is to use a remote sensing dataset and the Quantum Geographic Information System (QGIS) to map and assess active captive limestone mining sites at the Yerraguntla in the YSR Kadapa district, Andhra Pradesh, India, for the year 2019. In this paper, the Limestone mining area was assessed as 379.57 ha with an overall accuracy of 95.79%, user accuracy of 97.25%, producer accuracy of 99.18%, and kappa coefficient of 0.957 by analyzing Sentinel-2A imagery dataset with Random Forest (RF) classifier. The findings were assessed in 2019 using very high-resolution (<1m) Google Earth images (486.47 ha) from Google Earth Pro and industrial field survey reports (487.10 ha). This research helps Limestone mine owners and Environmental engineers implement eco-friendly mining operations and monitor mining progress at low cost with less human efforts for sustainable development.

The Nexus between Land Use/Cover changes and Land Surface Temperature: Remote sensing based Two-Decadal Analysis

The use of Remote Sensing (RS) data is crucial for promptly detecting and monitoring changes in both short and long term, providing real time information on Land Use/Cover (LULC), Land Surface Temperature (LST), and Enhanced Vegetation Index (EVI), adapting spatio-temporal variations. The primary focus of this study is to assess the effect of LULC changes on LST in Tashk-Bakhtegan and Maharloo (TBM) lakes basin, Iran, within 2001, 2011, and 2021, using MODIS data. Specifically, five main LULC classes involving: water body, rangeland, cropland, urban area, and bareland were identified. Beside accuracy and transition of LULC maps using User Accuracy (UA), Producer Accuracy (PA), and Kappa Coefficient (KC), the analysis included changes in LULC, Enhanced Vegetation Index (EVI), and LST, as well as the relationship among them when vegetation cover was at its peak. Moreover, a one-way analysis of variance (ANOVA) test was performed to group these variables using Duncan's test. The results showed that the accuracy of LULC maps were more than 84% for all the years. Furthermore, the conversion of croplands to rangelands showed the most significant changes, with a total of 1311.38 km2 during 2001–2021. Average EVI remained almost stable across the total area, whereas average LST generally increased by 0.65 °C. Barelands consistently exhibited the highest temperatures in all the years, followed by urban areas. While no significant changes were observed in the EVI averages, significant changes were observed in the LST across all LULC classes in different years. The results also indicated a consistent negative correlation between LST and EVI, stronger in croplands than rangelands, with Spearman's correlation coefficient of −0.714, −0.674, and −0.623 over the total area in 2001, 2011, and 2021, respectively. The findings are crucial for land planners to comprehend the effects of LULC changes on LST to adopt appropriate strategies in the TBM lakes basin.

Machine learning for urban land use/ cover mapping: Comparison of artificial neural network, random forest and support vector machine, a case study of Dilla town

The ability to accurately classify land use/cover (LULC) is critical for environmental monitoring and land use planning. This study compares three machine learning algorithms: Artificial Neural Network (ANN), Support Vector Machine (SVM), and Random Forest (RF) for LULC classification using Google Earth images from the years 2006, 2014, and 2022. The objective of this study is to evaluate and identify the best classifier for LULC classification and change detection. Four LULC categories (Built-up, Open area, Farmland, and Agroforestry) were identified. The evaluation criteria included overall accuracy, kappa coefficient, producer's accuracy, user's accuracy, computing time, algorithm stability, and visual quality. The results showed that the RF algorithm outperformed both SVM and ANN algorithms with an average overall accuracy of 0.97, kappa coefficient of 0.98, producer's accuracy of 0.99, and user's accuracy of 0.97, surpassing the accuracies achieved by SVM (0.96, 0.97, 0.98, and 0.97) and ANN (0.89, 0.81, 0.94, and 0.88), with corresponding computing times of 6.33, 15, and 30 s. All classifiers performed stably with different training sizes. Visual quality assessment revealed that RF had the highest precision. Consequently, the built-up change detection result shows, the net change in built-up area between 2006 and 2022 was increased by 0.74 Km2for ANN, 1.74 Km2 for SVM, and 1.66 Km2for RF. The comparison reveals that the RF algorithm showcasing high precision in detecting change, consistent with the data (increased by 1.65 Km2) obtained from Dilla town land administration office. To validate the results, the study considered field surveys, reference images, local experts, and previous studies. Based on the findings, the study concludes that using RF classifier with an object-based approach is an effective way to map LULC and detect changes in the study area over time. Future researchers are recommended to utilize this effective algorithm for addressing LULC related problems in the study area.

Automatic pine wilt disease detection based on improved YOLOv8 UAV multispectral imagery

The pine wilt disease (PWD) can cause destructive death to pine trees in a short period. Utilizing unmanned aerial vehicle (UAV) remote sensing technology to promptly identify PWD-infected trees has become an effective and feasible method for precise PWD monitoring. In this study, UAV multispectral imagery was used to analyze the sensitive spectral bands and different vegetation indices for PWD discriminability. A dataset of optimal spectral combinations from visible light and multispectral images was constructed, along with an improved YOLOv8 deep learning model for rapid and accurate identification of PWD-infected trees. The improved YOLOv8 model used omni-dimensional dynamic convolution (ODConv) to enhance the performance of convolutional networks, designed a dynamic head (DyHead) module to capture PWD features more accurately, and applied MPDioU to improve the regression accuracy and model runtime efficiency. Experimental results showed that the mAP@0.5 of the improved YOLOv8 model increased to 89.1 %, with a user accuracy of 90 % and a recall rate of 93.1 %. This achieved rapid and accurate detection of PWD-infected trees, providing effective technical support for automatic identification of PWD epidemic areas and control of PWD outbreaks based on UAV multispectral imagery.

Pemetaan Otomatis Air Permukaan dengan Pendekatan Random Forest dan Citra Sentinel-2

Understanding the presence and distribution of the Earth's surface water is crucial for various environmental planning segments such as water resources management, natural disaster mitigation, environmental conservation, and spatial planning. In addition, the availability of Sentinel-2 satellite imagery data that offers high spatial resolution with wide coverage allows for more accurate and detailed surface water mapping compared to traditional methods. This research presents the utilization of the Random Forest algorithm as a surface water mapping model tested on Sentinel-2 image data and can be applied nationally with a spatial resolution of 10 meters. In this case, the Random Forest model was developed using spectral indices such as Modified Normalized Difference Water Index (MNDWI) and Normalized Vegetation Index (NDVI) as well as true color (RGB) spectral bands. The accuracy assessment results show that the Random Forest model built can perform surface water mapping very accurately, which is confirmed by the evaluation metrics in the form of overall accuracy, average producer's accuracy, and average user's accuracy which have a value of 98.12%, 98.13%, and 98.07%, respectively. Based on these results, it can be seen that the model has significant potential to assist the environmental planning process further through surface water mapping.

A U-Net Model for Urban Land Cover Classification Using VHR Satellite Images

Urban settings are dynamic, constantly changing, and presenting a wide range of surface materials with high diversity in both spatial and spectral variation. As a result, mapping urban growth, evaluating infrastructure, managing water resources, and monitoring natural land cover become more complex tasks. Urban applications have made considerable progress thanks to the abundance of VHR orbital data and the recent development of artificial intelligence strategies especially neural networks. Convolutional neural networks have the potential to significantly enhance the analysis of urban land cover by addressing the limitations of traditional techniques. U-Net is a popular neural network for land cover analysis in remote sensing images. The current research presents a CNN model employing U-Net for image semantic segmentation in urban study area using both spectral and spatial context of VHR satellite data. The proposed model is trained, validated, and tested for VHR satellite image classification into five urban classes: water, vegetation, bare soil, road, and building. The CNN semantic segmentation results are compared to maximum likelihood image classification outcomes for validation and stability evaluation. A confusion matrix is applied to the classified scenes to determine the overall accuracy, producer's and user's accuracy, and Kappa coefficient using 400 random points with their corresponding ground truth. The U-Net image semantic segmentation technique achieved an overall accuracy of 87.50% and Kappa coefficient of 0.8395 which outperforms the maximum likelihood classification method with an overall accuracy of 83.25% and Kappa coefficient of 0.7812.

Research on the design of growable solid wood children’s beds

To meet children’s needs for beds at different stages of growth, a product design model integrating the KANO Model, the Hierarchical Analysis Method, and Axiomatic Design (KANO-AHP-AD) was adopted to conduct a design study on the growability of solid wood children’s beds. In accordance with the KANO model, the demands for children’s beds coming from questionnaires among families with children were classified and sorted, and the demand indicators were then summarized. Secondly, AHP was introduced to establish a multilevel hierarchical model, construct a judgment matrix of design elements, and calculate the weights of these elements to improve the accuracy of users’ demand weights. Then, AD was used to complete the mapping of the demand domain, function domain, and design domain of children’s beds and to judge their reasonableness through the matrix. Such a design allows users to evaluate and verify the rationality of the program. Through modular design techniques, the growability requirements were fulfilled. A low number of product modules were freely combined to form product types with multiple functions to meet customers’ needs for personalization and functional diversification. The work has value for the design of growable children’s beds, thus contributing to sustainable development and environmental protection causes.

Leveraging machine learning for analyzing the nexus between land use and land cover change, land surface temperature and biophysical indices in an eco-sensitive region of Brahmani-Dwarka interfluve

This present study aims to evaluate land use and land cover changes using five machine-learning algorithms in Google Earth Engine. The performance of these machine learning algorithms was evaluated using user accuracy, producer accuracy, overall accuracy, and kappa coefficient. Additionally, it seeks to understand the evolving pattern of land surface temperature and its correlation with biophysical characteristics in the Brahmani-Dwarka interfluve region from 1991 to 2021. The results demonstrate that RF algorithms outperform other algorithms in terms of performance, with RF algorithms averaging an 86 % overall accuracy and a Kappa coefficient of 0.82, while GTB comes in second with an 85 % overall accuracy and a Kappa value of 0.81. SVM performed moderately, while CART and MD algorithms struggled to perform in this study. The analysis of land transformation from 1991 to 2021 indicates that the stone crushing industry, built-up, and waterbodies have shown an upward trend, while vegetation and fallow land have decreased in their geographical extent. The results of the LST analysis indicated the study area had an LST rise of 8.72 °C over the last 30 years, or 0.29 °C per year, with the stone crushing and mining activities exhibiting the highest increase of 11.67 °C. Further, correlation analysis reveals LST has a highly significant positive correlation with NDBI and NDBaI and a very significant negative correlation with NDVI, MNDWI, and NDLI throughout the study period. The findings emphasize the importance of long-term planning and environmentally friendly development, ensuring responsible stone crushing activities, sustainable techniques, biodiversity conservation, and sustainable utilization of natural resources.

Identifying long-term burned forests in the rugged terrain of Southwest China:A novel method based on remote sensing and ecological mechanisms

Burned forests were detected using remote sensing techniques. Yet, identifying long-term burned forests in the mountains, especially at a large spatial extent, remains a great challenge due to a lack of long-term high-resolution remote sensing data or the unsatisfactory performance of the moderate-resolution remotely sensed data in complex mountain landscapes. Efficient, robust, and cost-effective methods are urgently called for. In this study, we developed a novel method combining remote sensing and ecological mechanisms (short for RSEM-M) with a paired-pixel approach to identify the long-term burned forests in Southwest China, where the terrain is rugged. In mapping burned forests in 2010, the overall accuracy, producer accuracy, user accuracy, Kappa coefficient, and field validation accuracy were 92.27 %, 95.86 %, 88.36 %, 0.85, and 81.16 %, respectively. Compared to the standard procedure and FireCCI51 and MCD64A1 fire products, the RSEM-M shows higher accuracy and consistency in identifying burned forests of different sizes, with fewer invalid pixels, better spatial continuity, and more accurate boundary delineation. Then, using the RSEM-M, we mapped burned forests in Southwest China from 2002 to 2017. The cumulative areas were 37.69 × 104 ha. The RSEM-M improved the efficiency and accuracy of long-term burned forest identifications in mountainous landscapes at a large spatial extent.