{"id":{"repo_id":"unsw","oai_identifier":"oai:unsworks.library.unsw.edu.au:1959.4/64895"},"canonical_url":"https://search.dev.ndltd.org/etd/unsw/oai:unsworks.library.unsw.edu.au:1959.4/64895","repository":{"repo_id":"unsw","name":"University of New South Wales","base_url":"https://unsworks.unsw.edu.au/oai/provider"},"display":{"title":"Quantitative rice mapping with remote sensing image time series","abstract":"This thesis focuses on developing a chain of machine learning algorithms for quantitative rice mapping with remote sensing time series, including data pre-processing, crop identification, rice classification, and rice variety separation. Time-series optical and Synthetic-Aperture Radar (SAR) data delivered by the Sentinel satellites are utilised. Testing sites are at a major rice planting area in southwest New South Wales, Australia. A superpixel-based multi-kernel selection approach is proposed to correct the angular effects in multi-temporal remote sensing images. This correction is performed adaptively for different land cover types. In this way the unique bidirectional reflection property of each land cover type is accommodated. To separate cropping lands from other land cover types, an efficient sequential classifier training approach, SCT-SVM, is developed. This approach sequentially classifies time-series images in a cost-efficient manner. For the training of each new image, a priori knowledge from previous images is leveraged, thus considerably reducing training data requirement. A temporal-adaptive support vector machine, TA-SVM, is designed to transfer knowledge from previous images to the new image. The TA-SVM outperforms two other state-of-the-art algorithms (i.e., A-SVM and PMT-SVM). An optical-SAR-synergistic rice mapping approach is proposed to distinguish rice from non-rice crops, where a nomination-favoured opinion pool, NF-OP, is designed to synergistically combine the optical and SAR data. This is an improvement to the classic veto-prone logarithmic opinion pool to alleviate the miss-detection problem of direct drilling rice, which is induced by the reduced usage of irrigation water. It is challenging to classify different rice varieties. A deep spectral-temporal convolutional neural network, deep-CNN, is then built. The deep-CNN is directed to learn fine features in both spectral and temporal domains, aiming to extract the unique spectral characteristics and growing phenology of each rice variety. A large quantity of ground reference data is used for network training, with superior performance being observed in comparison with traditional classifiers. The proposed algorithms and experimental results presented in this thesis demonstrate the effectiveness of applying remote sensing time series to quantitative rice mapping. They provide promising supports to the Australian rice industry for refining farm management practices in future.","abstract_html":"This thesis focuses on developing a chain of machine learning algorithms for quantitative rice mapping with remote sensing time series, including data pre-processing, crop identification, rice classification, and rice variety separation. Time-series optical and Synthetic-Aperture Radar (SAR) data delivered by the Sentinel satellites are utilised. Testing sites are at a major rice planting area in southwest New South Wales, Australia. A superpixel-based multi-kernel selection approach is proposed to correct the angular effects in multi-temporal remote sensing images. This correction is performed adaptively for different land cover types. In this way the unique bidirectional reflection property of each land cover type is accommodated. To separate cropping lands from other land cover types, an efficient sequential classifier training approach, SCT-SVM, is developed. This approach sequentially classifies time-series images in a cost-efficient manner. For the training of each new image, a priori knowledge from previous images is leveraged, thus considerably reducing training data requirement. A temporal-adaptive support vector machine, TA-SVM, is designed to transfer knowledge from previous images to the new image. The TA-SVM outperforms two other state-of-the-art algorithms (i.e., A-SVM and PMT-SVM). An optical-SAR-synergistic rice mapping approach is proposed to distinguish rice from non-rice crops, where a nomination-favoured opinion pool, NF-OP, is designed to synergistically combine the optical and SAR data. This is an improvement to the classic veto-prone logarithmic opinion pool to alleviate the miss-detection problem of direct drilling rice, which is induced by the reduced usage of irrigation water. It is challenging to classify different rice varieties. A deep spectral-temporal convolutional neural network, deep-CNN, is then built. The deep-CNN is directed to learn fine features in both spectral and temporal domains, aiming to extract the unique spectral characteristics and growing phenology of each rice variety. A large quantity of ground reference data is used for network training, with superior performance being observed in comparison with traditional classifiers. The proposed algorithms and experimental results presented in this thesis demonstrate the effectiveness of applying remote sensing time series to quantitative rice mapping. They provide promising supports to the Australian rice industry for refining farm management practices in future.","abstract_has_math":false,"creators":["Guo, Yiqing"],"institution":"UNSW, Sydney","degree_name":null,"degree_level":null,"degree_discipline":null,"degree_department":null,"school":null,"contributors":[],"advisors":[],"committee_chairs":[],"committee_members":[],"year":2019,"date_issued":"2019","date_published":"2019","updated_at":"2026-07-24T05:33:43Z","subjects":["Machine learning","Remote sensing","Rice mapping"],"languages":["EN"],"rights":["open access","CC BY-NC-ND 3.0","free_to_read"],"rights_urls":["https://purl.org/coar/access_right/c_abf2","https://creativecommons.org/licenses/by-nc-nd/3.0/au/"],"identifier_entries":[{"key":"dc:identifier","label":"Identifier","values":["https://doi.org/10.26190/unsworks/21585"],"render_values":[{"text":"https://doi.org/10.26190/unsworks/21585","href":"https://doi.org/10.26190/unsworks/21585","code":true}]}]},"links":{"outbound_url":"http://hdl.handle.net/1959.4/64895","outbound_label":"Handle","outbound_source":"dc:identifier"},"metadata_groups":[{"id":"people","label":"People","entries":[{"key":"dc:creator","label":"Author","values":["Guo, Yiqing"]}]},{"id":"academic_context","label":"Academic Context","entries":[{"key":"dc:date","label":"Dc Date","values":["2019"]},{"key":"dc:publisher","label":"Institution","values":["UNSW, Sydney"]},{"key":"dc:type","label":"Dc Type","values":["doctoral thesis","http://purl.org/coar/resource_type/c_db06"]}]},{"id":"subjects_keywords","label":"Subjects and Keywords","entries":[{"key":"dc:subject","label":"Dc Subject","values":["Machine learning","Remote sensing","Rice mapping"]}]},{"id":"language_rights","label":"Language and Rights","entries":[{"key":"dc:language","label":"Dc Language","values":["EN"]},{"key":"dc:rights","label":"Dc Rights","values":["open access","https://purl.org/coar/access_right/c_abf2","CC BY-NC-ND 3.0","https://creativecommons.org/licenses/by-nc-nd/3.0/au/","free_to_read"]}]},{"id":"identifiers","label":"Identifiers","entries":[{"key":"dc:identifier","label":"Identifier","values":["http://hdl.handle.net/1959.4/64895","https://unsworks.unsw.edu.au/bitstreams/1a02dd58-f1c8-4430-9ec0-a0f3454ef48c/download","https://doi.org/10.26190/unsworks/21585"]}]},{"id":"additional","label":"Additional Metadata","entries":[{"key":"dc:description","label":"Description","values":["This thesis focuses on developing a chain of machine learning algorithms for quantitative rice mapping with remote sensing time series, including data pre-processing, crop identification, rice classification, and rice variety separation. Time-series optical and Synthetic-Aperture Radar (SAR) data delivered by the Sentinel satellites are utilised. Testing sites are at a major rice planting area in southwest New South Wales, Australia. A superpixel-based multi-kernel selection approach is proposed to correct the angular effects in multi-temporal remote sensing images. This correction is performed adaptively for different land cover types. In this way the unique bidirectional reflection property of each land cover type is accommodated. To separate cropping lands from other land cover types, an efficient sequential classifier training approach, SCT-SVM, is developed. This approach sequentially classifies time-series images in a cost-efficient manner. For the training of each new image, a priori knowledge from previous images is leveraged, thus considerably reducing training data requirement. A temporal-adaptive support vector machine, TA-SVM, is designed to transfer knowledge from previous images to the new image. The TA-SVM outperforms two other state-of-the-art algorithms (i.e., A-SVM and PMT-SVM). An optical-SAR-synergistic rice mapping approach is proposed to distinguish rice from non-rice crops, where a nomination-favoured opinion pool, NF-OP, is designed to synergistically combine the optical and SAR data. This is an improvement to the classic veto-prone logarithmic opinion pool to alleviate the miss-detection problem of direct drilling rice, which is induced by the reduced usage of irrigation water. It is challenging to classify different rice varieties. A deep spectral-temporal convolutional neural network, deep-CNN, is then built. The deep-CNN is directed to learn fine features in both spectral and temporal domains, aiming to extract the unique spectral characteristics and growing phenology of each rice variety. A large quantity of ground reference data is used for network training, with superior performance being observed in comparison with traditional classifiers. The proposed algorithms and experimental results presented in this thesis demonstrate the effectiveness of applying remote sensing time series to quantitative rice mapping. They provide promising supports to the Australian rice industry for refining farm management practices in future."]},{"key":"dc:format","label":"Dc Format","values":["application/pdf"]},{"key":"dc:title","label":"Title","values":["Quantitative rice mapping with remote sensing image time series"]}]}],"canonical_facts":{"dc:creator":["Guo, Yiqing"],"dc:date":["2019"],"dc:description":["This thesis focuses on developing a chain of machine learning algorithms for quantitative rice mapping with remote sensing time series, including data pre-processing, crop identification, rice classification, and rice variety separation. Time-series optical and Synthetic-Aperture Radar (SAR) data delivered by the Sentinel satellites are utilised. Testing sites are at a major rice planting area in southwest New South Wales, Australia. A superpixel-based multi-kernel selection approach is proposed to correct the angular effects in multi-temporal remote sensing images. This correction is performed adaptively for different land cover types. In this way the unique bidirectional reflection property of each land cover type is accommodated. To separate cropping lands from other land cover types, an efficient sequential classifier training approach, SCT-SVM, is developed. This approach sequentially classifies time-series images in a cost-efficient manner. For the training of each new image, a priori knowledge from previous images is leveraged, thus considerably reducing training data requirement. A temporal-adaptive support vector machine, TA-SVM, is designed to transfer knowledge from previous images to the new image. The TA-SVM outperforms two other state-of-the-art algorithms (i.e., A-SVM and PMT-SVM). An optical-SAR-synergistic rice mapping approach is proposed to distinguish rice from non-rice crops, where a nomination-favoured opinion pool, NF-OP, is designed to synergistically combine the optical and SAR data. This is an improvement to the classic veto-prone logarithmic opinion pool to alleviate the miss-detection problem of direct drilling rice, which is induced by the reduced usage of irrigation water. It is challenging to classify different rice varieties. A deep spectral-temporal convolutional neural network, deep-CNN, is then built. The deep-CNN is directed to learn fine features in both spectral and temporal domains, aiming to extract the unique spectral characteristics and growing phenology of each rice variety. A large quantity of ground reference data is used for network training, with superior performance being observed in comparison with traditional classifiers. The proposed algorithms and experimental results presented in this thesis demonstrate the effectiveness of applying remote sensing time series to quantitative rice mapping. They provide promising supports to the Australian rice industry for refining farm management practices in future."],"dc:format":["application/pdf"],"dc:identifier":["http://hdl.handle.net/1959.4/64895","https://unsworks.unsw.edu.au/bitstreams/1a02dd58-f1c8-4430-9ec0-a0f3454ef48c/download","https://doi.org/10.26190/unsworks/21585"],"dc:language":["EN"],"dc:publisher":["UNSW, Sydney"],"dc:rights":["open access","https://purl.org/coar/access_right/c_abf2","CC BY-NC-ND 3.0","https://creativecommons.org/licenses/by-nc-nd/3.0/au/","free_to_read"],"dc:subject":["Machine learning","Remote sensing","Rice mapping"],"dc:title":["Quantitative rice mapping with remote sensing image time series"],"dc:type":["doctoral thesis","http://purl.org/coar/resource_type/c_db06"]},"updated_at":"2026-07-24T05:33:43Z"}