[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"doc-detail-126910-en":3,"doc-seo-126910-105":30,"detail-sidebar-cat-0-en-105":90},{"code":4,"msg":5,"data":6},0,"success",{"doc_id":7,"user_id":8,"nickname":9,"user_avatar":10,"doc_module":4,"category_id":11,"category_name":12,"doc_title":13,"doc_description":14,"doc_content":15,"file_id":16,"file_url":17,"file_type":18,"file_size":19,"view_count":4,"is_deleted":4,"is_public":20,"is_downloadable":20,"audit_status":20,"page_count":21,"language":22,"language_code":23,"site_id":24,"html_lang":23,"table_of_contents":25,"faqs":26,"seo_title":27,"seo_description":14,"update_tm":28,"read_time":29},126910,2336474459895,"Aria","https://ap-avatar.wpscdn.com/avatar/22000baeef7a5ed0655?x-image-process=image/resize,m_fixed,w_180,h_180&k=1786071322749376916",8,"Research & Report","Exploring CaSSIS with Machine Learning - The Search for Dust Devils on Mars","Machine learning is used to systematically map dust devils in the CaSSIS (Colour and Stereo Surface Imaging System) image dataset. Dust devils are atmospheric vortices whose spatial and temporal behavior informs Mars atmospheric dynamics, yet existing orbital image scans are limited. A Yolov5x detector is trained on CaSSIS-labeled instances, then deployed in a streaming pipeline for calibrated map-projected NPB composites. The model finds 255 dust devils, revealing strong seasonal and local-noon occurrence patterns and estimating typical velocities from stereo timing.","54th Lunar and Planetary Science Conference 2023 (LPI Contrib. No. 2806) 1479. pdf  \nExploring CaSSIS with Machine Learning – The Search for Dust Devils on Mars. V. T. Bickel 1, S. J.  \nConway2, N. Thomas3, M. Read3, A. Valantinas3, E. Hauber4, P. Grindrod5, J. Wray6, V. Rangarajan7, and the CaSSIS Science Team,1Center for Space and Habitability, University of Bern, CH ([valentin.bickel@unibe.ch](valentin.bickel@unibe.ch)), 2University of Nantes, FR, 3University of Bern, CH, 4German Aerospace Center, GER, 5Natural History Museum, UK, 6Georgia Institute of Technology, USA, 7University of Western Ontario, CAN.  \nIntroduction: Dust devils are atmospheric vortices made visible by dust entrained from the martian surface [1] . Their spatial and temporal distribution, shape, and velocity provide insights into Mars’atmospheric dynamics, useful for, e.g., dust, weather and climate modelling [1,2] . Yet, most of the available orbital image datasets have not been thoroughly scanned for dust devils and/or are limited to images acquired at a fixed local time (sun-synchronous orbits), limiting the representativeness of current analyses. Here, we use machine learning to systematically map dust devils in the CaSSIS (Colour and Stereo Surface Imaging System) image dataset. CaSSIS is a color and stereo imaging system onboard ESA’s Trace Gas Orbiter [3] (on a non-sun-synchronous orbit) with a nominal spatial resolution of 4.6 m, covering ~6 % of Mars’ surface as of November 2022 (33,130 images) .  \nMethods: We collate all dust devil instances currently known to the CaSSIS science team (n=62) . Each instance is annotated with a rectangular bounding box, referred to as labels. All labels are split into a training (n=57) and validation set (n=5) and used to tune a COCO pre-trained convolutional neural network called Yolov5x (PyTorch 1.7) . In total, we train the neural net over 250 epochs (t=~10 minutes), while applying ample label augmentation, including rotation, translation, scaling, and radiometric modifications (brightness, hue, etc.) . We further include negative training (i.e., non-dust devil sites, n=44) and validation images (n=6) to reduce false positives during inference (deployment) . The neural net achieves a mean average precision of 0.95 in the validation set. We note that both the training and validation set are extremely small – the validation performance is therefore unlikely to be representative of the inference performance.  \nWe deploy the best iteration of the neural net in a pre-existing processing pipeline [4] that was modified to stream and process calibrated, map-projected CaSSIS NPB composites (NIR – near-infrared, PAN panchromatic, BLU -blue) . In NPB composites, dust devils feature a bright core, an adjacent shadow, and distinct color fringes that are caused by their movement between the acquisitions of the individual color channels (Fig. 1) . The processing of the entire map-projected NPB dataset took ~24 h (~1041 images per hour) using a single work station with one NVIDIA RTX 3090, running on 4 individual threads.  \nResults: The neural net identified a total of 255 dust devils (Fig. 1) . Most of the detected dust devils were observed at Ls ~= 130° and ~270°, between 11 AM and 1 PM local time, and in Amazonis Planitia (Fig. 2 & 3) . The largest dust devils were observed during local noon, while their occurrence follows the seasons (Fig. 2 & 3) . A number of dust devils happen to be captured by CaSSIS stereo pairs: using the time difference between the individual acquisitions (~45 s) we derive an average dust devil velocity of~10 m/s for sites in Arabia Terra and Aonia Terra (Fig. 3) . We note that the overall ratio of dust devil/number of images scanned is one order of magnitude smaller in CaSSIS (0.01) than in CTX (0. 1) as found by earlier work [2] .  \nDiscussion: The abundance of dust devils around local noon agrees well with models and lander/rover observations [5], and our velocity estimates are in line with earli","cbCaiqpD2cCy4lw3","https://ap.wps.com/l/cbCaiqpD2cCy4lw3","pdf",526997,1,2,"English","en",105,"# Introduction\n# Methods\n# Results\n# Discussion\n# Future work","[{\"question\":\"Why are dust devils important for Mars studies, and what limitation exists in current datasets?\",\"answer\":\"Dust devils provide information about Mars atmospheric dynamics through their distribution, shape, and velocity. Many existing orbital image datasets were not thoroughly scanned and are often limited to fixed local time observations, reducing representativeness.\"},{\"question\":\"How was the dust devil detector trained and validated?\",\"answer\":\"All known CaSSIS dust devil instances were annotated with bounding boxes and split into training and validation sets. A COCO pre-trained Yolov5x network was fine-tuned with label augmentation, and negative training plus additional validation images were included to reduce false positives.\"},{\"question\":\"What were the key findings from deploying the model to CaSSIS images?\",\"answer\":\"The neural net identified 255 dust devils. Detections were concentrated around Ls ≈ 130° and ≈ 270° between 11 AM and 1 PM local time, with occurrences varying by season; stereo timing supported an estimated average velocity of about 10 m/s in specific regions.\"}]","Exploring CaSSIS with Machine Learning - The Search for Dust Devils on Mars | PDF",1785935593,5,{"code":4,"msg":31,"data":32},"ok",{"site_id":24,"language":23,"slug":33,"title":13,"keywords":34,"description":14,"schema_data":35,"social_meta":85,"head_meta":87,"extra_data":89,"updated_unix":28},"exploring-cassis-with-machine-learning-the-search-for-dust-devils-on-mars","",{"@graph":36,"@context":84},[37,53,67],{"@type":38,"itemListElement":39},"BreadcrumbList",[40,44,47,50],{"item":41,"name":42,"@type":43,"position":20},"https://docshare.wps.com","Home","ListItem",{"item":45,"name":46,"@type":43,"position":21},"https://docshare.wps.com/document/","Document",{"item":48,"name":12,"@type":43,"position":49},"https://docshare.wps.com/document/research-report/",3,{"item":51,"name":13,"@type":43,"position":52},"https://docshare.wps.com/document/exploring-cassis-with-machine-learning-the-search-for-dust-devils-on-mars/126910/",4,{"url":51,"name":13,"@type":54,"author":55,"headline":13,"publisher":57,"fileFormat":60,"inLanguage":23,"description":14,"dateModified":61,"datePublished":61,"encodingFormat":60,"isAccessibleForFree":62,"interactionStatistic":63},"DigitalDocument",{"name":9,"@type":56},"Person",{"url":41,"name":58,"@type":59},"DocShare","Organization","application/pdf","2026-08-05",true,{"@type":64,"interactionType":65,"userInteractionCount":4},"InteractionCounter",{"@type":66},"ViewAction",{"@type":68,"mainEntity":69},"FAQPage",[70,76,80],{"name":71,"@type":72,"acceptedAnswer":73},"Why are dust devils important for Mars studies, and what limitation exists in current datasets?","Question",{"text":74,"@type":75},"Dust devils provide information about Mars atmospheric dynamics through their distribution, shape, and velocity. Many existing orbital image datasets were not thoroughly scanned and are often limited to fixed local time observations, reducing representativeness.","Answer",{"name":77,"@type":72,"acceptedAnswer":78},"How was the dust devil detector trained and validated?",{"text":79,"@type":75},"All known CaSSIS dust devil instances were annotated with bounding boxes and split into training and validation sets. A COCO pre-trained Yolov5x network was fine-tuned with label augmentation, and negative training plus additional validation images were included to reduce false positives.",{"name":81,"@type":72,"acceptedAnswer":82},"What were the key findings from deploying the model to CaSSIS images?",{"text":83,"@type":75},"The neural net identified 255 dust devils. Detections were concentrated around Ls ≈ 130° and ≈ 270° between 11 AM and 1 PM local time, with occurrences varying by season; stereo timing supported an estimated average velocity of about 10 m/s in specific regions.","https://schema.org",{"og:url":51,"og:type":86,"og:title":13,"og:site_name":58,"og:description":14},"article",{"robots":88,"canonical":51},"index,follow",{"doc_id":7,"site_id":24},{"code":4,"msg":5,"data":91},[92,96,100,104,108,113,118,121,126,129,133],{"id":20,"doc_module":4,"doc_module_name":46,"category_name":93,"show_sort_weight":94,"slug":95},"Story & Novel",90,"story-novel",{"id":21,"doc_module":4,"doc_module_name":46,"category_name":97,"show_sort_weight":98,"slug":99},"Literature",80,"literature",{"id":52,"doc_module":4,"doc_module_name":46,"category_name":101,"show_sort_weight":102,"slug":103},"Exam",70,"exam",{"id":29,"doc_module":4,"doc_module_name":46,"category_name":105,"show_sort_weight":106,"slug":107},"Comic",60,"comic",{"id":109,"doc_module":4,"doc_module_name":46,"category_name":110,"show_sort_weight":111,"slug":112},6,"Technology",50,"technology",{"id":114,"doc_module":4,"doc_module_name":46,"category_name":115,"show_sort_weight":116,"slug":117},7,"Healthcare",40,"healthcare",{"id":11,"doc_module":4,"doc_module_name":46,"category_name":12,"show_sort_weight":119,"slug":120},30,"research-report",{"id":122,"doc_module":4,"doc_module_name":46,"category_name":123,"show_sort_weight":124,"slug":125},9,"Religion & Spirituality",20,"religion-spirituality",{"id":124,"doc_module":4,"doc_module_name":46,"category_name":127,"show_sort_weight":124,"slug":128},"World Cup","world-cup",{"id":130,"doc_module":4,"doc_module_name":46,"category_name":131,"show_sort_weight":130,"slug":132},10,"Lifestyle","lifestyle",{"id":134,"doc_module":4,"doc_module_name":46,"category_name":135,"show_sort_weight":29,"slug":136},19,"General","general"]