Iterative Self-Organizing Data Analysis Research Articles

Evolutive multi-attribute decision making with online consumer reviews

In the digital age, the sheer volume of online consumer reviews imposes a cognitive burden on consumers, complicating their purchasing decisions. Many studies have integrated consumer opinions to provide consumers with clear and concise information. However, these studies often prioritize mainstream opinions, overlooking the diversity and timeliness of other important perspectives. To address this challenge, we propose an evolutive decision-making method. Firstly, we propose an attribute rating evolution algorithm to address the online reviews based on the iterative self-organizing data analysis technique and time decay. This algorithm enables real-time analysis of the diverse opinions expressed in review data. Then, taking into account consumer attribute preferences and decision-making psychology, we formulate multiple product ranking strategies to offer personalized decisions based on the evolutive opinions. Our method decreases the bias towards review quantity, ensuring that significant opinions are not overshadowed by more frequent ones. Through data experiments and an application on OpenTable.com, we demonstrate that our method can provides effective decision recommendation for consumers.

Optimization of UWB indoor positioning based on hardware accelerated Fuzzy ISODATA

With the development of wireless communication technology, Ultra-Wideband (UWB) has become an important solution for indoor positioning. In complex indoor environments, the influence of non-line-of-sight (NLOS) factors leads to increased positioning errors. To improve the positioning accuracy, fuzzy iterative self-organizing data analysis clustering algorithm (ISODATA) is introduced to process a large amount of UWB data to reduce the influence of NLOS factors, and to stabilize positioning error within 2 cm, enhances the accuracy of the positioning system. To further improve the running efficiency of the algorithm, FPGA is used to accelerate the key computational part of the algorithm, compared with running on the MATLAB platform, which improves the speed about 100 times, enhances the algorithm’s computational speed dramatically.

A New and Robust Index for Water Body Extraction from Sentinel-2 Imagery

Land surface water is a key part in the global ecosystem balance and hydrological cycle. Remote sensing has become an effective tool for its spatio-temporal monitoring. However, remote sensing results exemplified in so-called water indices are subject to several limitations. This paper proposes a new and effective water index called the Sentinel Multi-Band Water Index (SMBWI) to extract water bodies in complex environments from Sentinel-2 satellite imagery. Individual tests explore the effectiveness of the SMBWI in eliminating interference of various special interfering cover features. The Iterative Self-Organizing Data Analysis Technique Algorithm (ISODATA) method and confusion matrix along with the derived accuracy evaluation indicators are used to provide a threshold reference when extracting water bodies and evaluate the accuracy of the water body extraction results, respectively. The SMBWI and eight other commonly used water indices are qualitatively and quantitatively compared through vision and accuracy evaluation indicators, respectively. Here, the SMBWI is proven to be the most effective at suppressing interference of buildings and their shadows, cultivated lands, vegetation, clouds and their shadows, alpine terrain with bare ground and glaciers when extracting water bodies. The overall accuracy in all tests was consistently greater than 96.5%. The SMBWI is proven to have a high ability to identify mixed pixels of water and non-water, with the lowest total error among nine water indices. Most notably, better results are obtained when extracting water bodies under interfering environments of cover features. Therefore, we propose that our novel and robust water index, the SMBWI, is ready to be used for mapping land surface water with high accuracy.

Accident black spot clustering oriented maritime search and rescue resource allocation and optimization

Efficient and rapid deployment of maritime search and rescue(MSAR) resources is a prerequisite for maritime emergency search and rescue, in order to improve the efficiency and accuracy of MSAR. This paper proposes an integrated approach for emergency resource allocation. The approach encompasses three main steps: identifying accident black spots, assessing high-risk areas, and optimizing the outcomes through a synergistic combination of an optimization algorithm and reinforcement learning. In the initial step, the paper introduces the iterative self-organizing data analysis technology (ISODATA) for identifying accident spots at sea. A comparative analysis is conducted with other clustering algorithms, highlighting the superiority of ISODATA in effectively conducting dense clustering. This can effectively carry out dense clustering, instead of the situation where the data spots are too dispersed or obvious anomalies that affect the clustering. Furthermore, this approach incorporates entropy weighting to reassess the significance of accident spots by considering both the distance and the frequency of accidents. This integrated approach enhances the allocation of search and rescue forces, ensuring more efficient resource utilization. To address the MSAR vessel scheduling problem at sea, the paper employs the non-dominated sorting genetic algorithm II combined with reinforcement learning (NSGAII-RL). Comparative evaluations against other optimization algorithms reveal that the proposed approach can save a minimum of 7% in search and rescue time, leading to enhanced stability and improved efficiency in large-scale MSAR operations. Overall, the integrated approach presented in this paper offers a robust solution to the ship scheduling problem in maritime search and rescue operations. Its effectiveness is demonstrated through improved resource allocation, enhanced timeliness, and higher efficiency in responding to maritime accidents.

An assessment of seasonal changes in forests using the segmentation and cluster analysis of Landsat space images

For the purposes of monitoring the state of forest ecosystems, it is most effective to use the capabilities of remote methods. On multispectral Landsat satellite images (a time series of summer and autumn images for 2000, 2001, 2013, 2014) of the territory of the Olekminsky State Nature Reserve, fragments of forests dominated by Gmelin larch (Larix gmelinii Rupr.) with an area of 250 km2 (scale 1:5000). Then the polygons were saved at three levels of segmentation (scaling) – 4, 16, 64 with scales of 1:2500, 1:1250, 1:625. During decryption, uncontrolled classification of polygons using the ISODATA (Iterative Self-Organizing Data Analysis Technigue) method into 2,4,10 classes was carried out. Classification into two classes was used to calculate the forest cover index. Seasonal changes were determined by the difference in the forest cover index values of the polygons in summer and autumn. It has been shown that the greater the difference, the greater the proportion of larch in mixed forest stands. Distribution curves of forest cover index values were constructed for polygons of the 3rd scaling level. Based on the results of classification into 4, 10 classes, statistical processing was carried out with the calculation of indicators of difference and similarity of polygons - dispersion of the general population and Fisher's test (F-test). The results of changes in population dispersion and F-test at different levels of segmentation and in different years are considered.

A study of high-resolution remote sensing image landslide detection with optimized anchor boxes and edge enhancement

ABSTRACTThis paper takes landslide as a special research object. For the problems of landslide detection in remote sensing images, deep learning and playback method is adopted. Using the You Only Look Once v5 network (YOLOv5) in combination with the Gabor filter, its edge detection, detection anchor frame and small object detection scale are improved and optimized. The YOLOv5(ISODATA) model was finally established for landslide image detection by incorporating the edge control factor and four clustering algorithms (K-means, K-means + +, k-medoid, and Iterative Self-Organizing Data Analysis Techniques Algorithm (ISODATA) to evaluate the accuracy of the detection anchor frame and add small target large-scale sampling. Three target identification models – YOLOv5, Region Convolution Neural Network (R-CNN), and Fast R-CNN – are experimentally compared in order to assess the effectiveness of the proposed method. According to the results of experiments, the proposed method outperforms the other three detection models with an AUC of 0.921, a recall of 86.14%, and an MCC of 0.887. It further demonstrates the method’s positive impact on landslide remote sensing image recognition and its ability to solve related issues.

Assessment of open-pit captive limestone mining areas using sentinel-2 imagery with spectral indices and machine learning algorithms

Limestone mining is a significant economic activity in India, accounting for around 10% of the GDP however, it has certain negative environmental consequences. The objective of this study is to determine the spatial distribution area of captive limestone mines using remote sensing datasets, spectral index, and machine learning algorithms and compare their area estimation with industrial field survey reports for the financial year 2019. The study area includes a limestone resource area of 2226.16 ha with an excavation area of 487.10 ha in 2019. In the present research, we used a high-resolution Sentinel-2A satellite dataset to map and compute the active mining area by implementing the Normalised Vegetation Index (NDVI), Iterative Self-Organizing Data Analysis Technique (ISODATA), K-Nearest Neighbours (KNN), and Random Forest (RF) algorithms in the QGIS 3.18 software tool. The RF classifier estimated a limestone mine area of 379.57 ha with user accuracy (UA) of 97.25% and producer accuracy (PA) of 99.18% with a kappa coefficient value of 0.957. The mine area was estimated at 417.47 ha with a UA of 98.99% and PA of 99.10% and kappa value of 0.947 of the KNN method, The NDVI method estimated 469.92 ha with a UA of 93.63% and PA of 92.04% and kappa value 0.685. This research confirmed that the RF classifier well performed in classification with overall accuracy (OA) of 95.79% to KNN (OA of 94.78%), NDVI (OA of 79.84%) classifiers, and ISODATA poor in classification with OA of 64.16%. This research assists limestone mine owners and environmental engineers in making environmentally sustainable decisions, eco-friendly mine design, and monitoring.

UAV-based seagrass wrack orthophotos classification for estimating blue carbon

Seagrass wrack, the detritus from seagrass meadows that accumulates along the coast, is thought to contribute to carbon sequestration in shallow coastal areas when processed into fixed biochar. Accurate and efficient estimation methods for seagrass wrack area size and carbon content are still being explored, especially across multiple seasons. The carbon content of seagrass wrack along the Futtsu coast of Tokyo Bay, Japan, was evaluated using unmanned aerial vehicle (drone) photogrammetry and in-situ measurements of volume and density. A combination of a low-pass filter in pre-processing and the unsupervised iterative self-organizing data analysis technique algorithm (ISODATA) classification method achieved the best classification accuracy when compared with a majority filter and/or any of the three conventional supervised classifiers (support vector machine, random forest, and maximum likelihood classifier). This approach minimizes the manual labor required for determining the area size of the seagrass wrack. The Futtsu tidal flat seagrass wrack carbon for 2020 was estimated to be 3 tons C/yr, accounting for approximately 2% of the seagrass meadow net primary production. These results can greatly assist blue carbon policy implementation.

Day-Ahead Dynamic Assessment of Consumption Service Reserve Based on Morphological Filter

With the development goal of a low-cost and low-carbon reserve market, this paper proposes a dynamic assessment method for day-ahead consumption service reserve demand considering the forecast error of uncertainty power. The iterative self-organizing data analysis techniques algorithm is adopted to cluster the historical actual power into typical scenarios. In addition, the online matching between the typical scenario and the day-ahead forecast power is conducted. In order to realize the hierarchical quantification of reserve demand, the reserve resources in the whole power system are classified according to their response time. Furthermore, the mathematical morphology filter based on the structural elements that are consistent with the response time of the hierarchical reserve resources is initially applied to decompose the historical forecast error of the matched scenarios. The simulation results verify that the proposed dynamic assessment effectively reduces the reserve cost on the basis of being able to cope with multi-time-scale power fluctuations.

Reconstruction of spatially continuous time-series land subsidence based on PS-InSAR and improved MLS-SVR in Beijing Plain area

ABSTRACT Beijing has undergone severe settlement in recent years. Persistent Scatterers Interferometric Synthetic Aperture Radar (PS-InSAR) technique has been widely used to derive time-series land deformation. However, existing studies have faced two challenges: (1) the nonlinear characteristics of time-series subsidence has not been fully investigated; (2) since PS points are normally distributed in urban areas with high building density, measurement gaps usually exist in nonurban areas. To address the challenges, we presented a new method to reconstruct spatially continuous time-series deformation. First, PS-InSAR was used to retrieve the deformation based on 135 scenes of Envisat ASAR and Radarsat-2 images from 2003 to 2020. Polynomial Curve Fitting (PCF) was then used to model nonlinear time-series deformation for the PS points. In the PS measurement gaps, Iterative Self-Organizing Data Analysis Technique (ISODATA) and Multi-output Least Squares Support Vector Regression (MLS-SVR) were used to estimate the PCF coefficients and then time-series deformation considering 40 features including thickness of the compressible layers, annual groundwater level, etc. The major results showed that (1) compared to linear, quadratic, and quartic models, cubic polynomial model generated better fit for the time-series deformation (R2 ≈0.99), suggesting obvious nonlinear temporal pattern of deformation; (2) the time-series deformation over measurement gaps reconstructed by ISODATA and MLS-SVR had satisfactory accuracy (R2 = 0.92, MAPE < 15%) and yielded higher accuracy (R2 = 0.947) than IDW (R2 = 0.687) and Ordinary Kriging (R2 = 0.688) interpolation methods. The reconstructed results maintain the nonlinear characteristics and ensure the high spatial resolution (120 m) of time-series deformation. Among the 40 predictor variables, ground water level datasets are the most influential predictors of time-series deformation.

A New Method for Continuous Track Monitoring in Regions of Differential Land Subsidence Rate Using the Integration of PS-InSAR and SBAS-InSAR

It is difficult for single time-series Interferometric Synthetic Aperture Radar (InSAR) processing to guarantee the accuracy and efficiency of continuous track monitoring in regions of differential subsidence. This paper proposes a new method, integrating the Persistent Scatterer InSAR (PS-InSAR) with high precision and the Small Baseline Subset InSAR (SBAS-InSAR) with high efficiency for continuous track monitoring in regions of differential land subsidence rates. Based on PS-InSAR processing, the Iterative Self-Organizing Data Analysis Techniques (ISODATA) algorithm is adopted to search the boundary of differential subsidence between slow and fast subsidence rates. The SBAS-InSAR processing with high frequency is used to continuously track and monitor the regions with fast subsidence rates incorporating original data and newly added data into small data sets from time to time according to SAR data updating, the monitoring results of which are obtained from the weighted average of the added results of SBAS-InSAR processing and the original results of PS-InSAR processing. The impact of SAR data updating on the slow subsidence rate region is so weak that it is not necessary to simultaneously update the corresponding monitoring results to improve global efficiency. If the slow subsidence rates region must be remeasured in relation to its previous subsidence, or the proportion of new data capacity alters compared with the original data set, PS-InSAR processing is used to analyze the whole monitoring region again using the complete data set. A case study performed on the west region of the Qinhuai River in Nanjing, China, indicates that the density of monitoring points in the fast-subsidence region is greatly improved, increasing from 711 points/km2 to 2760 points/km2—an increase of 288.2%.

An assessment of forest restoration after fie using segmentation and cluster analysis of Landsat images

For the purposes of monitoring the state of forest ecosystems, it is most effctive to use the capabilities of remote methods. Fragments with an area of 250 km2 (scale 1:5000) were identifid on Landsat multispectral satellite images (time series of summer images for 1995, 2000, 2004, 2008, 2013, 2016) of the Olyokminsky State Nature Reserve. Then polygons were saved at three levels of segmentation: 4, 16, 64 with the scales of 1:2500, 1:1250, 1:625. When deciphering, an unmanaged classifiation of polygons was carried out using the ISODATA method (Iterative Self-Organizing Data Analysis Technigue) into 2, 4, 10 classes. Distribution curves for the values of the forest cover index for polygons of the 3rd level of segmentation were constructed. The results of classifiation into 4 classes were used to calculate the thematic pixel diffrence. According to the classification results for 4.10 classes, statistical processing was carried out with the calculation of the diffrence and similarity indicators of polygons, the dispersion of the general population and the Fisher test (F-test). A method is described for determining the disturbance of ecosystems and their restoration from the distribution curves of the forest cover index. The results of the change in the dispersion of the general population and the F-test at diffrent levels of segmentation and at diffrent stages of forest restoration are considered. The features of the transition between three levels of self-similarity (scaling) of multifractal structures as forests are restored are determined.

Analysis of Spatial and Temporal Variation in Water Coverage in the Sub-Lakes of Poyang Lake Based on Multi-Source Remote Sensing

As the largest freshwater lake in China, Poyang Lake is an internationally important wetland and the largest migratory bird habitat in Asia. Many sub-lakes distributed in the lake basin are seasonal lakes, which have a significant impact on hydro-ecological processes and are susceptible to various changes. In this study, using multi-source remote sensing data, a continuous time-series construction method of water coverage suitable in Poyang Lake was developed. That method combined the downscaling of the MNDWI (modified normalized difference water index) with the ISODATA (iterative self-organizing data analysis technique algorithm), and its accuracy can be up to 97% in the months when Landsat 8 is available or 87% when it is unavailable. Based on that method, the increasing variation in water coverage was observed in the sub-lakes of Poyang Lake during 2013–2020 to be within a range of 200–690 km2 normally. The center of the sub-lakes always remained inundated (&gt;80% inundation frequency), while the surrounding areas were probably kept dry for seven months (except for June to September). The dominant influencing factors of water coverage variations were different in different hydrological periods (wet season and dry–wet season: discharge; dry season: temperature and wind speed; wet–dry season: temperature and precipitation). In addition, “returning farmland to lakes” affected the increase in the water area in the sub-lakes. This study is helpful for the management of water resources and the protection of migratory birds in the Poyang Lake region.

Potential of DESIS and PRISMA hyperspectral remote sensing data in rock classification and mineral identification:a case study for Banswara in Rajasthan, India.

Remote sensing datasets and methods are suitable for mapping and managing the natural resources like minerals, clean water, and energy and also govern their sustainability nowadays. Hyperspectral (HS) imaging has immense potential for rock type classification, mineral mapping, and identification. This work demonstrates the potential of feature extraction techniques and unsupervised machine learning methods for the space-borne hyperspectral remote sensing data in characterizing and identifying mineral and classifying rock type in Banswara, Rajasthan, India. Feature extraction techniques can reveal variations within the data, which can help identify geological areas, reduce noise, and check the dimensionality of the data. Singular value decomposition (SVD)-based principal component analysis (PCA), kernel PCA (KPCA), minimum noise fraction (MNF), and independent component analysis (ICA) were tested for lithological mapping using recently launched DLR Earth Sensing Imaging Spectrometer Hyperspectral (DESIS) and PRecursore IperSpettrale della Missione Applicativa (PRISMA) data in order to map geologically significant areas. Unsupervised machine learning methods, such as Iterative Self-Organizing Data Analysis Technique (ISODATA) and K-means, were also employed. Vertex component analysis (VCA) was utilized to check for similarity and identify various spectral features. Our work demonstrates the advantages of using feature extraction algorithms such as PCA and KPCA over MNF and ICA in geological mapping and interpretability. We recommend K-means as the preferred method for lithological classification of hyperspectral remote sensing data. Our work highlights the potential of advanced feature extraction algorithms for mineral mapping using hyperspectral data, providing different ways to identify minerals and ultimately leading to better mineral resource management.

Sophisticated CPBIS methods applied for FBISODATA clustering algorithm using with real time image database

Data mining is a process of mining hidden information to the previously unknown data and theoretically useful unknown information from a large amount of genuine data to be stored in a database. Image mining is a part of data mining with used as a predictive measure to identify with the age of the tiger. This research work is mainly focused on, to identify with the age of the Tiger using data mining techniques. This research work incorporates with which those domains of image processing and data mining to predict the age of the tiger using different kinds of color images are used. The fuzzy iterative self-organizing data analysis (FISODATA) clustering method requires more predefined parameters tofind the maximum number of iterations, the minimum number of points in the cluster, and smallest amount of distance with the centers of the clusters. The key undertaking of the studies of diverse colors mechanism is to decide the age of the tiger; the usage of shade action pixel primarily based on image segmentation; the usage of facts that are used in the mining techniques. However, the more matrix components to be measuring the processing time, retrieval time, accuracy, and blunders fee with the aid of using producing better performance.

A Novel Curve Pattern Recognition Framework for Hot-Rolling Slab Camber

Camber is a typical asymmetrical defect of slabs in the hot-rolling process. The identification of camber plays a key role in improving the quality of the finished strip. To obtain the shape information of the camber, we select the derivative dynamic time warping (DDTW) distance as the measure of curve similarity. An iterative self-organizing data analysis clustering algorithm integrated with DDTW is proposed to divide the sample into different clusters. A curve template is generated via polynomial fitting in each cluster based on the dynamic mechanism of camber. Subsequently, the dynamic time warping (DTW) and DDTW distances are adopted as the input and a weighted random forest (WRF) algorithm is developed for the camber classification to address the sample imbalance problem. Comparative experiments are conducted on the camber sample collected in the actual factory to test the clustering and classification performances. The experimental results indicate that the error rates (ERRs) of four distance-based methods─Euclidean distance, Pearson distance, DTW, and DDTW are 32.7%, 28.5%, 19.1% and 12.3%, respectively, and ERRs of the four classification algorithms—DDTW-SVM, DDTW-KNN, DDTW-RF, and DDTW-WRF are 9.8%, 5.7%, 4.7%, and 2.1%, respectively, which verifies the superior performance of the proposed method in this article.

Identification of geochemical anomalies related to mineralization: A case study from porphyry copper deposits in the Qulong-Jiama mining district of Tibet, China

Geochemical exploration data plays a vital role in mineral prospectivity mapping (MPM) for discovering unknown mineral deposits. In this study, compositional balance analysis (CoBA), unsupervised and supervised learning are collaboratively used to improve the practice of identification of geochemical anomalies related to mineralization in the Qulong-Jiama mining district of Tibet. For CoBA, five balances of geochemical elements were constructed by the sequential binary partition (SBP) technique, facilitating the interpretation of geochemical/geological processes and contrasting with geochemical anomalies generated by unsupervised and supervised learning. The iterative self-organizing data analysis techniques algorithm (ISODATA) and isolation forest (iForest) are used for preprocessing the geochemical data and therefore optimizing the appropriate training sets (e.g., positive and negative training samples) for supervised learning. Different training datasets including locations of known mineralization, randomly selected positive samples from resulting clusters by ISODATA and outliers by iForest are fed to the support vector machine (SVM) and random forests (RF) algorithm. The results of high AUC values indicate that supervised learning can effectively delineate prospective areas both with SVM and RF, however, the excessive large area of high-prospectivity and a bias toward known mineralization makes the outcomes of MPM unpractical. Nevertheless, a hybrid of the CoBA, unsupervised learning and supervised learning could alleviate such situations and provide insights into MPM using the geochemical data.

Effectiveness of the red-edge band of RapidEye in land cover classification

ABSTRACT This study examined the effectiveness of the red-edge band using RapidEye satellite images for land cover classification. The analysis comprises three schemes for evaluating the effectiveness of the red-edge band: principal component analysis (PCA), vegetation index, and supervised image classification. The factor loadings computed by means of PCA were applied to analyze the importance of each band in the training samples. The analysis results of the factor loadings indicated that the red-edge band performed better than the visible band in the vegetation region. When rice paddy and peanuts were classified using the NDVI_RE, the improvement in accuracy was approximately 7%. Further, the accuracy of rice paddy classification using CMFI_RE was improved by approximately 6%. It can thus be inferred that the red-edge band made a certain contribution to vegetation classification. In land cover classification using reflectance, the accuracy of the support vector machine (SVM) was higher than that of the maximum likelihood classifier (MLC), the iterative self-organizing data analysis technique, and the K-means algorithm. When the red-edge band was included, the overall accuracy improved from 1% to 3%. The results of our experiments indicated that the red-edge band contributed marginally to land cover classification.

Research on Grape-Planting Structure Perception Method Based on Unmanned Aerial Vehicle Multispectral Images in the Field

In order to accurately obtain the distribution of large-field grape-planting sites and their planting information in complex environments, the unmanned aerial vehicle (UAV) multispectral image semantic segmentation model based on improved DeepLabV3+ is used to solve the problem that large-field grapes in complex environments are affected by factors such as scattered planting sites and complex background environment of planting sites, which makes the identification of planting areas less accurate and more difficult to manage. In this paper, firstly, the standard deviation (SD) and interband correlation of UAV multispectral images were calculated to obtain the best band combinations for large-field grape images, and five preferred texture features and two preferred vegetation indices were screened using color space transformation and grayscale coevolution matrix. Then, supervised classification methods, such as maximum likelihood (ML), random forest (RF), and support vector machine (SVM), unsupervised classification methods, such as the Iterative Self-organizing Data Analysis Techniques Algorithm (ISO DATA) model and an improved DeepLabV3+ model, are used to evaluate the accuracy of each model in combination with the field visual translation results to obtain the best classification model. Finally, the effectiveness of the classification features on the best model is verified. The results showed that among the four machine learning methods, SVM obtained the best overall classification accuracy of the model; the DeepLabV3+ deep learning scheme based on spectral information + texture + vegetation index + digital surface model (DSM) obtained the best accuracy of overall accuracy (OA) and frequency weight intersection over union (FW-IOU) of 87.48% and 83.23%, respectively, and the grape plantation area relative error of extraction was 1.9%. This collection scheme provides a research basis for accurate interpretation of the planting structure of large-field grapes.

Planet Four: A Neural Network’s search for polar spring-time fans on Mars

Dark deposits visible from orbit appear in the Martian south polar region during the springtime. These are thought to form from explosive jets of carbon dioxide gas breaking through the thawing seasonal ice cap, carrying dust and dirt which is then deposited onto the ice as dark ‘blotches’, or blown by the surface winds into streaks or ‘fans’. We investigate machine learning (ML) methods for automatically identifying these seasonal features in High Resolution Imaging Science Experiment (HiRISE) satellite imagery. We designed deep Convolutional Neural Networks (CNNs) that were trained and tested using the catalog generated by Planet Four, an online citizen science project mapping the south polar seasonal deposits. We validated the CNNs by comparing their results with those of ISODATA (Iterative Self-Organizing Data Analysis Technique) clustering and as expected, the CNNs were significantly better at predicting the results found by Planet Four, in both the area of predicted seasonal deposits and in delineating their boundaries. We found neither the CNNs or ISODATA were suited to predicting the source point and directions of seasonal fans, which is a strength of the citizen science approach. The CNNs showed good agreement with Planet Four in cross-validation metrics and detected some seasonal deposits in the HiRISE images missed in the Planet Four catalog; the total area of seasonal deposits predicted by the CNNs was 27% larger than that of the Planet Four catalog, but this aspect varied considerably on a per-image basis.