{"about":{"site":"https://codewithpapers.app","non_affiliation":"Code with Papers and Syntology are not affiliated with, endorsed by, or sponsored by Papers with Code, Meta, or the pwc-archive mirror.","licence":"CC BY-SA 4.0","licence_url":"https://creativecommons.org/licenses/by-sa/4.0/legalcode","attribution":"https://codewithpapers.app/attribution","modified":"archive material modified by Syntology; see the attribution page"},"url":"/paper/spectral-spatial-classification-of-1","title":"Spectral-spatial classification of hyperspectral images: three tricks and a new supervised learning setting","arxiv_id":"1711.05512","date":"2017-11-15","proceeding":null,"authors":["Jacopo Acquarelli","Elena Marchiori","Lutgarde M. C. Buydens","Thanh Tran","Twan van Laarhoven"],"abstract":"Spectral-spatial classification of hyperspectral images has been the subject\nof many studies in recent years. In the presence of only very few labeled\npixels, this task becomes challenging. In this paper we address the following\ntwo research questions: 1) Can a simple neural network with just a single\nhidden layer achieve state of the art performance in the presence of few\nlabeled pixels? 2) How is the performance of hyperspectral image classification\nmethods affected when using disjoint train and test sets? We give a positive\nanswer to the first question by using three tricks within a very basic shallow\nConvolutional Neural Network (CNN) architecture: a tailored loss function, and\nsmooth- and label-based data augmentation. The tailored loss function enforces\nthat neighborhood wavelengths have similar contributions to the features\ngenerated during training. A new label-based technique here proposed favors\nselection of pixels in smaller classes, which is beneficial in the presence of\nvery few labeled pixels and skewed class distributions. To address the second\nquestion, we introduce a new sampling procedure to generate disjoint train and\ntest set. Then the train set is used to obtain the CNN model, which is then\napplied to pixels in the test set to estimate their labels. We assess the\nefficacy of the simple neural network method on five publicly available\nhyperspectral images. On these images our method significantly outperforms\nconsidered baselines. Notably, with just 1% of labeled pixels per class, on\nthese datasets our method achieves an accuracy that goes from 86.42%\n(challenging dataset) to 99.52% (easy dataset). Furthermore we show that the\nsimple neural network method improves over other baselines in the new\nchallenging supervised setting. Our analysis substantiates the highly\nbeneficial effect of using the entire image (so train and test data) for\nconstructing a model.","url_abs":"http://arxiv.org/abs/1711.05512v4","url_pdf":"http://arxiv.org/pdf/1711.05512v4.pdf","source":{"archive":"pwc-archive (Hugging Face), CC BY-SA 4.0","snapshot":"2025-07-28","licence_url":"https://creativecommons.org/licenses/by-sa/4.0/legalcode","row_kind":"abstracts"},"code_links":[{"paper_slug":"spectral-spatial-classification-of-1","repo_url":"https://bitbucket.org/TeslaH2O/cnn_hyperspectral","is_official":1,"mentioned_in_paper":1,"mentioned_in_github":0,"framework":"none","reach":null}],"tasks":[{"task_slug":"data-augmentation","task_name":"Data Augmentation"},{"task_slug":"classification","task_name":"General Classification"},{"task_slug":"hyperspectral-image-classification","task_name":"Hyperspectral Image Classification"},{"task_slug":"image-classification","task_name":"Image Classification"},{"task_slug":"image-classification","task_name":"image-classification"}],"methods":[],"datasets_introduced":[],"methods_introduced":[],"results":[],"syntology":{"syntology_url":null,"atlas_url":null,"mcp":null,"developers":"https://syntology.ai/developers"},"arxiv_metadata":null,"syntology_extracted_results":null}