Satellite Image Classification Research Articles

Holdout based blending approaches for improved satellite image classification

An essential component of remote sensing, image analysis, and pattern recognition is image categorization. The classification of land use using remotely sensed data creates a map-like representation as the final form of the investigation. With its ability to effectively categorize satellite images, machine learning (ML) algorithms have gained significant traction in a number of fields, including land-use planning, disaster response, and natural resource management. Ensemble learning is also a widely used technique for enhancing the precision of satellite image categorization, which combines multiple models to get more precise predictions. Holdout is an ensemble technique, where multiple ML algorithms are used for training on the same dataset. The primary goal of this study is to create a holdout model for classifying satellite images. Initially, this study explores the usage of ML algorithms namely support vector machines (SVM), k-nearest neighbor (KNN), decision trees (DT), gradient boosting classifier (GBC), histogram-based GBC (HGBC), random forest classifier (RF), bagging classifier (BC), XGBoost classifier for classifying satellite images. Later, GBC, HGBC, RF, BC, and XGBoost are combined to build a stacking model. The bagging ensemble model outperforms all other methods and reaches an accuracy of 88.90%. Finally, blending models with holdout approach were developed and achieved accuracy of 93.70%, 94.14%, and 93.87% which outperformed all previous algorithms.

Design of a High-Efficiency Temporal Engine for Real-Time Spatial Satellite Image Classification Using Augmented Incremental Transfer Learning for Crop Analysis

Design of a High-Efficiency Temporal Engine for Real-Time Spatial Satellite Image Classification Using Augmented Incremental Transfer Learning for Crop Analysis

Implementation of satelite image classification using PyTorch

This article presents a practical implementation of the satellite image classification task using the PyTorch deep learning library. The aim of the work is to develop a neural network model capable of classifying satellite images into four classes: water, sands, clouds, and greenery. The model is trained using the pre-trained ResNet architecture. The proposed approach allows for automatic classification of satellite images, fa-cilitating the analysis of large amounts of data, and has been tested on Sentinel-2 satellite images acquired from March to August 2023 for parts of Kherson and Mykolaiv regions, Ukraine, and achieved an average accuracy of 95%. The presented research demonstrates the practical use of modern deep learning methods, in particular the PyTorch library, to solve actual problems of satellite image analysis and classification.

Optimal trained ensemble of classification model for satellite image classification

Optimal trained ensemble of classification model for satellite image classification

Few-shot satellite image classification for bringing deep learning on board OPS-SAT

Bringing artificial intelligence on board Earth observation satellites unlocks unprecedented possibilities to extract actionable items from various image modalities at the global scale in real time. This is of paramount importance nowadays, as downlinking large amounts of imagery is not only prohibitively expensive but also time-consuming. However, building deep learning solutions that could be deployed on board an edge device is challenging due to the limited manually-annotated satellite datasets and hardware constraints of an edge device. This paper addresses these challenges through harnessing a blend of data-centric and model-centric approaches to build a well-generalizing yet efficient and resource-frugal deep learning model for multi-class satellite image classification in the few-shot learning settings. This integrated strategy is formulated to enhance classification performance while accommodating the unique demands of an image analysis chain on board OPS-SAT, a nanosatellite operated by the European Space Agency. The experiments performed over a real-world dataset of OPS-SAT images delves into the interactions between data- and model-centric techniques, underscores the significance of synthesizing artificial training data and emphasizes the value of ensemble learning. However, they also caution against negative transfer in domain adaptation. This study sheds light on effective model training strategies and highlights the multifaceted challenges inherent in deep learning for practical Earth observation, contributing insights to the field of satellite image classification within the constraints of nanosatellite operations. To ensure reproducibility of our study, the implementation is available at https://github.com/ShendoxParadox/Few-shot-satellite-image-classification-OPS-SAT.

DETECTING OBJECTS ON SATELLITE IMAGES USING PYTORCH

This paper is devoted to the development of a neural network for classifying satellite images of the Earth into four classes: water, sands, clouds, and green areas (forests, fields, and plant clusters). For this purpose, the Python library PyTorch is used, which is focused on the application of deep learning for computer vision tasks. The proposed approach allows for automatic classification of satellite images, facilitating the analysis of large amounts of data, and has been tested on Sentinel-2 satellite images acquired from March to August 2023 for some parts of Kherson and Mykolaiv regions, Ukraine, and achieved an overall accuracy of 95%. The results of the study are useful for specialists in the fields of remote sensing, computer vision, machine learning, and aerospace image analysis software development.

Detection of Land Use Changes in Forests Using Satellite Image Classification Based on Deep Learning: A Case Study of Sardasht Forests

Awareness of changes in forest areas has always been one of the most crucial environmental considerations globally. Sardasht forests is located in the northern Zagros region and with an area of around 91,117 hectares constitute for 24% of Iran total forests. They play an important role in water supply, soil preservation, climate regulation and overall economic and social balance in the country. Nevertheless, these forests are currently perceived as degraded due to the primary reason of tree cutting for fuel and livestock feed. In this study, remote sensing data of the study area including Landsat 5 TM satellite images from 2002, Landsat 7 ETM images from 2012, Landsat 8 OLI images 2020 and classification algorithms, maximum likelihood and artificial neural network were utilized. To examine the accuracy of each classification, ground truth points were randomly collected using GPS devices and by implementing control points, statistical parameters of error accuracy including Kappa coefficient and overall accuracy were calculated. The results show that the map generated using the artificial neural network algorithm has an overall accuracy of 98.32% with a Kappa coefficient of 0.9781 while the map produced by the maximum likelihood algorithm has an overall accuracy of 92.45% with a Kappa coefficient of 0.901. As a result, the artificial neural network algorithm demonstrates to be a more suitable method for producing land cover maps compared to the maximum likelihood algorithm.

Satellite image classification using deep learning approach

Satellite image classification using deep learning approach

Application of Machine Learning Methods for Satellite Image Classification: a Literature Review and Overview of Key Frameworks

Application of Machine Learning Methods for Satellite Image Classification: a Literature Review and Overview of Key Frameworks

Remote Sensing Image Segmentation for Aircraft Recognition Using U-Net as Deep Learning Architecture

Recognizing aircraft automatically by using satellite images has different applications in both the civil and military sectors. However, due to the complexity and variety of the foreground and background of the analyzed images, it remains challenging to obtain a suitable representation of aircraft for identification. Many studies and solutions have been presented in the literature, but only a few studies have suggested handling the issue using semantic image segmentation techniques due to the lack of publicly labeled datasets. With the advancement of CNNs, researchers have presented some CNN architectures, such as U-Net, which has the ability to obtain very good performance using a small training dataset. The U-Net architecture has received much attention for segmenting 2D and 3D biomedical images and has been recognized to be highly successful for pixel-wise satellite image classification. In this paper, we propose a binary image segmentation model to recognize aircraft by exploiting and adopting the U-Net architecture for remote sensing satellite images. The proposed model does not require a significant amount of labeled data and alleviates the need for manual aircraft feature extraction. The public dense labeling remote sensing dataset is used to perform the experiments and measure the robustness and performance of the proposed model. The mean IoU and pixel accuracy are adopted as metrics to assess the obtained results. The results in the testing dataset indicate that the proposed model can achieve a 95.08% mean IoU and a pixel accuracy of 98.24%.

The OPS-SAT case: A data-centric competition for onboard satellite image classification

AbstractWhile novel artificial intelligence and machine learning techniques are evolving and disrupting established terrestrial technologies at an unprecedented speed, their adaptation onboard satellites is seemingly lagging. A major hindrance in this regard is the need for high-quality annotated data for training such systems, which makes the development process of machine learning solutions costly, time-consuming, and inefficient. This paper presents “the OPS-SAT case”, a novel data-centric competition that seeks to address these challenges. The powerful computational capabilities of the European Space Agency’s OPS-SAT satellite are utilized to showcase the design of machine learning systems for space by using only the small amount of available labeled data, relying on the widely adopted and freely available open-source software. The generation of a suitable dataset, design and evaluation of a public data-centric competition, and results of an onboard experimental campaign by using the competition winners’ machine learning model directly on OPS-SAT are detailed. The results indicate that adoption of open standards and deployment of advanced data augmentation techniques can retrieve meaningful onboard results comparatively quickly, simplifying and expediting an otherwise prolonged development period.

A High-Efficiency Satellite Image Classification Method for Real-Time Application Using Augmented Incremental Transfer Learning

Researchers have put forth a range of models for processing satellite images, each with distinct data formats and processing requirements. One limitation arises from variations in the forms of modules, such as the image capturing and feature extraction modules, impacting accuracy and scalability in real-time scenarios. This passage introduces and explores a pioneering very competent temporal engine crafted specifically for classification of real-time satellite images. The proposed approach utilizes augmented incremental transfer learning, aiming to mitigate the limitations associated with diverse processing requirements. This approach involves capturing real-time satellite data through Google’s Earth Engine and subsequently processing it using a Convolutional Neural Network (CNN) based on transfer learning. The CNN employs backscatter coefficient analysis, utilizing coefficients derived from Precision Image's average intensity value across a distributed target. By integrating incremental learning and CNN for classification, the model achieves an impressive average accuracy of 98.06% in detecting crop type and severity of damage. Comparative analysis with state-of-the-art approaches reveals the superiority of the proposed model. It outperforms existing models by 5% in accuracy, showcasing its efficacy in satellite image processing and classification.

An improved post-hurricane building damaged detection method based on transfer learning

After a natural disaster, it is very important for the government to conduct a damaged assessment as soon as possible. Fast and accurate disaster assessment helps the government disaster relief departments allocate resources and respond quickly and effectively to minimize the losses caused by the disaster. Usually, the method of measuring disaster losses is to rely on manual field exploration and measurement, and then calculate and label the damaged buildings or land, or rely on unmanned collections to remotely collect pictures of the disaster-stricken area, and compare the original pictures to carry out the disaster annotation and calculation. These methods are time-consuming, labor-intensive, and inefficient. This paper proposes a post-hurricane building damage detection method based on transfer learning, which uses deep learning image classification algorithms to achieve post-disaster satellite image damage detection and classification, thereby improving disaster assessment efficiency and preparing for disaster relief and post-disaster reconstruction. The proposed method adopts the theory of transfer learning, establishes a disaster image detection model based on the convolutional neural network model, and uses the 2017 Hurricane Harvey data as the experimental data set. Experiments have proved that our proposed model accuracy of disaster detection reaches 97%, which is 1% higher than other models.

Comparison of Crop Area Estimation for the Winter Season Using the Google Earth Engine (GEE) Platform and Traditional Method of Supervisor Classification

There is a significant development in the methods and techniques for classifying satellite images during these years. One of the most recently launched methods is using the Google Earth Engine platform for its contribution in providing a huge amount of satellite data with different dates and multiple satellite images sources, with a method of work that is easy to deal with, but it needed to learn the computer language such as Java and Python, which is the method of work depends on writing a script that gives application commands to classify the different satellite image. Therefore, this research aims to apply classification of satellite images using the Google Earth Engine platform for three models Classification and Regression Trees (CART), Support Vector Machine (SVM), and Random Forest (RF), then compare the results calculated from using images classification (Supervised classification: Maximum likelihood) in terms of the degree of accuracy of the classification and matching the results of the classified satellite images in addition. to boundary clarify of the classification methods. This study showed that, using the Google Earth Engine (GEE) platform is considered one of the important methods that facilitated researchers interested in using remote sensing techniques and geographic information systems, due to the availability of this platform of free satellite sources with multiple spatial resolutions, in addition to the presence of a powerful system that collects, processes, and classifies satellite images using different classification models and indexes. But we need to apply these modern techniques at a number of levels, such as at Markas, Governorates, International, and regional levels, to judge their accuracy and suitability for large areas.

تصنيف الانسكابات النفطية على أساس الصور الفضائية باستخدام تقنيات التعلم العميق

التسرب النفطي هو تسرب في خطوط الأنابيب أو السفن أو منصات النفط أو الناقلات يؤدي إلى انطلاق المنتجات البترولية في البيئة البحرية أو على اليابسة بشكل طبيعي أو بسبب عمل بشري، مما يؤدي إلى أضرار جسيمة وخسائر مالية. تعد صور الأقمار الصناعية إحدى الأدوات القوية المستخدمة حاليًا لالتقاط المعلومات الحيوية والحصول عليها من سطح الأرض. لكن التعقيد والكم الهائل من البيانات يجعل من الصعب على البشر معالجتها ويستغرق وقتًا طويلاً. ومع ذلك، مع تقدم تقنيات التعلم العميق، أصبحت العمليات الآن محوسبة للعثور على المعلومات الحيوية باستخدام صور الأقمار الصناعية في الوقت الحقيقي. طبقت هذه الورقة ثلاث خوارزميات للتعلم العميق لتصنيف صور الأقمار الصناعية، بما في ذلك ResNet50، وVGG19، وInceptionV4؛ تم تدريبهم واختبارهم على مجموعة بيانات صور الأقمار الصناعية مفتوحة المصدر لتحليل كفاءة الخوارزميات وأدائها وربط دقة التصنيف والدقة والاستدعاء ودرجة f1. وأظهرت النتيجة أن InceptionV4 يعطي أفضل دقة تصنيف بنسبة 97% للغيوم والصحراوية والمناطق الخضراء والمياه، يليه VGG19 بنسبة 96% تقريبًا وResNet50 بنسبة 93%. أثبتت النتائج أن خوارزمية InceptionV4 مناسبة لتصنيف الانسكابات النفطية وعدم الانسكابات باستخدام صور الأقمار الصناعية على مجموعة بيانات تم التحقق من صحتها.

Satellite Image Representations for Quantum Classifiers

Existing quantum hardware is limited in the number of bits and length of the series of operations. Nevertheless, by shifting parts of the computation on classical hardware, hybrid quantum-classical systems utilize quantum hardware for scaled-down machine learning approaches, which is quantum machine learning. Due to the theoretically possible computational speed-up of quantum computers compared to classical computers and the increasing volume and speed of data generated in earth observation, attempts are now being made to use quantum computers for satellite image processing. However, satellite imagery is too large and high dimensional, and transformations that reduce the dimensionality are necessary to fit the classical data in the limited input domain of quantum circuits. This paper presents and compares several dimensionality reduction techniques as part of hybrid quantum-classical systems to represent satellite images with up to 256×256×3\\documentclass[12pt]{minimal} \\usepackage{amsmath} \\usepackage{wasysym} \\usepackage{amsfonts} \\usepackage{amssymb} \\usepackage{amsbsy} \\usepackage{mathrsfs} \\usepackage{upgreek} \\setlength{\\oddsidemargin}{-69pt} \\begin{document}$$256\ imes 256\ imes 3$$\\end{document} values with only 16 values. We evaluate the representations of two benchmark datasets with supervised classification by four different quantum circuit architectures. We demonstrate the potential use of quantum machine learning for satellite image classification and give a comprehensive overview of the impact of various satellite image representations on the performance of quantum classifiers. It shows that autoencoder models are best suited to create small-scale representations, outperforming commonly used methods such as principle component analysis.

Land use land cover change intensity analysis for sustainable natural resources management: The case of northwestern highlands of Ethiopia

Various studies conducted in the northwestern highlands of Ethiopia from 2001 to 2020 indicated two contrasting directions of change in land use land cover. One direction is the expansion of farmlands at the expense of the natural vegetation, and the other direction is expansion of plantation forest on farm and grazing lands. But, the net effect of land use land cover change is not as such clear. A clear understudying of such dynamics in land use land cover and the net effect of the change helps design effective land use planning policy, which enhances sustainable natural resources management. Accordingly, the study analyzed intensity of land use land cover change for three-time periods (1990–2000; 2000–2010 and 2010–2020). The findings of this study can be applicable to different parts of the globe where two contrasting directions of change in land use land cover exist. The study employed both qualitative and quantitative methods, which were specifically applied through carrying out step-by-step: satellite image acquisition, preprocessing, classification, accuracy assessment, and intensity analysis. Result of the analysis indicated that land use land cover change over the three intervals showed considerable variation where intensity was consistently increasing from the first (1990–2000) interval to the subsequent (2000–2010) interval, and it was fast (8.24%) during the third interval (2010–2020). The analysis also indicated considerably high transition from bare land to grazing land and bush land during the third interval, especially on areas where the community practiced controlled management of grazing lands (zero grazing) and protection of bush lands. From the total bare land cover in 2010, 21% converted to grazing land and 17% converted to bush land. LULC transition from bush land to cultivated land was rising due to encroachment of bush lands by farming. Transition from bush land to cultivated land increased by 4%, 16% and 50% in the period 1990–2000, 2000–2010, and 2010–2020, respectively. On the other hand, transition from cultivated land to forest land was considerably high during the third interval. From the total cultivated land cover in 2010, 25% converted to forest land, where the change is mainly associated to the introduction of woodlot plantation on individual farmland holdings. In general, the finding suggests that controlled management of grazing lands; comparing environmental and economic benefits of quarry/stone extraction sites; and plantation of economically viable and environmentally sound tree species are key policy issues which needs due attention in land use planning and management practices.

Spatial and Temporal Assessment of Land Use and Land Cover in Wilpattu National Park Restricted Zone, Sri Lanka


 
 
 The Wilpattu National Park (WNP) is the country's oldest and largest National Park. Aside from the direct destruction of plant and fauna ecosystems that sustain humans with vital resources, removing forest cover throws many crucial natural processes out of balance, which could have several detrimental effects. It is still unknown how the various management alternatives will affect the economic, ecological, and social elements of sustainable forest management and planning. Sustainable development and environmental protection are aided by wise land use and resource management. Many of the farmers reside in these marginal, underproductive forest communities. Through scenario analysis and using a forestry decision support system, this study evaluated the effects of several management approaches on crucial aspects of sustainable forest management. Geographical Information System (GIS), such as participatory online mapping, GPS mapping, and satellite image classification (Remote Sensing Technology), was used. The study employs an evaluation model for land use that considers biophysical and human-related variables as part of an integrated land use assessment. Based on the classification of the images, the environment's impact has been evaluated. Forty-nine Grama Niladhari (GN) divisions are in the influenced zone in the national park. Eight sample GN divisions were selected for detailed assessment based on the area's different landscapes and ecological and cultural settings. The current situation and encroachment were studied in each sample GN division. Also, Current threats, land use, and pressure activities were studied. The land use analysis of the Wilpattu restricted zone revealed the area of encroachment. Analyses conducted in the human village clearly show that the restricted area for human settlement is being encroached upon 2019, permanent building/housing units were found in 2003 satellite image categorization within the 1.6 km restricted zone. By 2021, the number mentioned above will be 3,810 instead. The past 20 years have seen an 89% growth in housing units. Land use analysis indicates nine zones of encroachment. Zones are dominant in encroachment in 19 GN Divisions in total. The Wilpattu forest village's Mullikulam and Eluwankulam GN Divisions are the most crucial and quickly encroaching areas. The highest percentages of restricted zone, land use analysis contain Paddy cultivation, and Chena cultivation shows the second highest value. Minimum land used is utilized for abandoned Chena cultivations. The highest percentage of land use agriculture is Kudawilachiya, Dematamalagama, and Nelumvila. In addition, the lowest agricultural land use was utilized in Horawila, Thalgaswewa, Helabawewa, Mirichchukaadi, and Adiyapualwakkulam.
 
 
 
 Keywords: Sustainable, Deforestation, Forest management, GIS, Restricted zone
 
 
 
 
 

Application of Geospatial Technology to Assess Pressures and Threats to Nature Related to Human Activities in Wilpattu National Park

Expanding agricultural land is one of the humanities largest and most considerable environmental impacts. It has transformed habitats and is one of the greatest pressures and threats to biodiversity. By 2020, with an estimated population of 22 million, the per capita arable land is about 0.158 hectares. This indicates heavy pressure on agricultural land use. This agricultural land use causes deforestation. Illegal deforestation results in the destruction of habitat and the restriction of wildlife movements. Apart from the direct destruction of habitats of flora and fauna that provide humans with valuable resources, removing forest cover puts many essential natural functions out of order, which may lead to a range of severe consequences. There have been many attempts at human encroachment at Wilpattu National Park, Sri Lanka, and its Forest Reserves; however, it has never been done at this scale with the awareness of the Government Authorities, who failed to act. The Easter end of the Wilpattu National Park lies between the divisional secretariat of Mahaviachchi, Karuwalagasweva, Nochchiyagama, and Rajanganaya. The buffer zone of 1.6km has been marked from the Wilpattu reserve boundary. This study uses the integrated land use assessment by applying a land use evaluation model that evaluates biophysical and human-related data in eight GN Divisions (Hunuvilagama, Kuda Vilachchiya, Track 13, Matha Kiramam, Palaikuli, Pahalapuliyankulam, Parana Eluwankulam, and Andiyapuliyankulam) in buffer zone in Wilpattu National Park. Environmental impact has been assessed based on image classification and area calculations. According to the analysis done in the human settlement, it indicates encroachment of the restricted zone for human settlement. The area of this buffer zone is approximately 33,147 ha. About 9,190 ha of this buffer zone area has been utilized by humans in various ways. Spatial- temporal data in 2003-2019 satellite image classification identified permanent building/housing units within the 1.6 km restricted zone, which has been increased in marginal villages. Mullikulam and Eluwankulam GN Divisions are the most critical and rapidly encroached areas in the forest village in the Wilpattu National Park. According to the GIS-based maps, cultivation, road construction, and residence are significant factors influencing the major ecosystems in the eastern part of the Wilpattu Park Boundary. Mahavilachchi and Nochchiyagama can be identified as the villages where elephants roam more and the areas where elephant-human conflict occurs regularly. Paddy and hay farming are primarily practiced in the areas close to the Wilpattu reserve zone. Additionally, coconut, banana, mango, and cashew crops are cultivated.
 
 Keywords: Wilpattu, Agriculture, Deforestation, Buffer zone, Human

A Comprehensive Review on Deep Learning Model for Multi Class Satellite Image Classification and Prediction

Abstract: Satellite images provide a wealth of information about the Earth's surface, but extracting meaningful insights from these vast datasets requires sophisticated analysis techniques. Deep learning models have emerged as powerful tools for multiclass satellite image classification and prediction, offering superior accuracy and efficiency compared to traditional methods. This comprehensive review delves into the latest advancements in deep learning architectures specifically designed for satellite image processing. It critically examines the performance of various convolutional neural network (CNN) architectures, exploring their strengths and limitations in handling diverse satellite image datasets. The review further investigates the integration of transfer learning and domain adaptation techniques to address challenges like limited training data and domain shift. Additionally, it sheds light on recent developments in interpretability and explain ability of deep learning models for satellite image analysis, enabling users to gain deeper understanding of the decision-making process behind predictions. By providing a holistic overview of the current landscape and future directions, this review serves as a valuable resource for researchers and practitioners working in the field of satellite image analysis and deep learning.