{"id":{"repo_id":"iastate","oai_identifier":"oai:dr.lib.iastate.edu:20.500.12876/nrQBl5Az"},"canonical_url":"https://search.dev.ndltd.org/etd/iastate/oai:dr.lib.iastate.edu:20.500.12876/nrQBl5Az","repository":{"repo_id":"iastate","name":"Iowa State University","base_url":"https://dr.lib.iastate.edu/server/oai/request"},"display":{"title":"Hierarchical transfer learning for small object detection","abstract":"Object detection faces a formidable challenge in detecting small objects due to their diminutive size and sparse pixel representation. This difficulty is exacerbated by the common lack of resolution in images, making the task of distinguishing small objects complex. Annotating data for training datasets also demands substantial manual effort and time, particularly when dealing with large numbers of small insects found in the field. Moreover, occlusion further complicates accurate detection, as small objects are prone to being obscured by other elements in the scene. To address these challenges, sophisticated techniques are needed, specifically tailored for small object detection within the broader field of object detection. In the context of insect detection, the task carries significant importance as certain insects pose a threat to agriculture and food security. Detecting these small pests in a timely manner is crucial to safeguarding food production and maintaining a stable agricultural economy. However, the insect detection task comes with various challenges, including the time-consuming annotation process for large datasets and difficulties in identifying small insects in low-resolution images or from a distance. Variations in insect life stages and similarities in shape and colors among different insect classes further add to the complexity. To overcome these challenges, this thesis explores and proposes two effective solutions: hierarchical transfer learning and slicing-aided hyper inference. Hierarchical transfer learning extends the concept of transfer learning by incorporating multiple steps of knowledge transfer from a large dataset to a target dataset. It leverages intermediate in-domain datasets to adapt the model progressively. On the other hand, slicing-aided hyper inference divides the original image into overlapping patches, enabling independent object detection on each patch and subsequent merging for accurate and comprehensive object detection outputs. The effectiveness of these proposed solutions was thoroughly evaluated through a comprehensive analysis, considering metrics such as mean Average Precision, Recall, and Precision. The evaluation focused on the accuracy of insect detection, especially in challenging scenarios. The thesis aims to contribute to more effective pest control measures, thereby safeguarding crops and supporting sustainable agricultural practices.","abstract_html":"Object detection faces a formidable challenge in detecting small objects due to their diminutive size and sparse pixel representation. This difficulty is exacerbated by the common lack of resolution in images, making the task of distinguishing small objects complex. Annotating data for training datasets also demands substantial manual effort and time, particularly when dealing with large numbers of small insects found in the field. Moreover, occlusion further complicates accurate detection, as small objects are prone to being obscured by other elements in the scene. To address these challenges, sophisticated techniques are needed, specifically tailored for small object detection within the broader field of object detection. In the context of insect detection, the task carries significant importance as certain insects pose a threat to agriculture and food security. Detecting these small pests in a timely manner is crucial to safeguarding food production and maintaining a stable agricultural economy. However, the insect detection task comes with various challenges, including the time-consuming annotation process for large datasets and difficulties in identifying small insects in low-resolution images or from a distance. Variations in insect life stages and similarities in shape and colors among different insect classes further add to the complexity. To overcome these challenges, this thesis explores and proposes two effective solutions: hierarchical transfer learning and slicing-aided hyper inference. Hierarchical transfer learning extends the concept of transfer learning by incorporating multiple steps of knowledge transfer from a large dataset to a target dataset. It leverages intermediate in-domain datasets to adapt the model progressively. On the other hand, slicing-aided hyper inference divides the original image into overlapping patches, enabling independent object detection on each patch and subsequent merging for accurate and comprehensive object detection outputs. The effectiveness of these proposed solutions was thoroughly evaluated through a comprehensive analysis, considering metrics such as mean Average Precision, Recall, and Precision. The evaluation focused on the accuracy of insect detection, especially in challenging scenarios. The thesis aims to contribute to more effective pest control measures, thereby safeguarding crops and supporting sustainable agricultural practices.","abstract_has_math":false,"creators":["Fotouhi, Fateme"],"institution":"Iowa State University","degree_name":"Master of Science","degree_level":"thesis","degree_discipline":"Computer science","degree_department":"Department of Computer Science","school":null,"contributors":[],"advisors":["Tian, Jin","Sarkar, Soumik","Singh, Arti"],"committee_chairs":[],"committee_members":[],"year":2023,"date_issued":"2023-12","date_published":"2023-12","updated_at":"2026-07-24T02:37:49Z","subjects":[],"languages":["en"],"rights":[],"rights_urls":[],"identifier_entries":[{"key":"dc:identifier.doi","label":"DOI","values":["https://doi.org/10.31274/td-20240617-205"],"render_values":[{"text":"https://doi.org/10.31274/td-20240617-205","href":"https://doi.org/10.31274/td-20240617-205","code":true}]}]},"links":{"outbound_url":"https://dr.lib.iastate.edu/handle/20.500.12876/nrQBl5Az","outbound_label":"Repository record","outbound_source":"dc:identifier.uri"},"metadata_groups":[{"id":"people","label":"People","entries":[{"key":"dc:contributor.advisor","label":"Advisor","values":["Tian, Jin","Sarkar, Soumik","Singh, Arti"]},{"key":"dc:contributor.department","label":"Department","values":["Department of Computer Science"]},{"key":"dc:creator","label":"Author","values":["Fotouhi, Fateme"]}]},{"id":"academic_context","label":"Academic Context","entries":[{"key":"dc:date.accessioned","label":"Dc Date Accessioned","values":["2024-01-25T20:17:39Z"]},{"key":"dc:date.available","label":"Dc Date Available","values":["2024-01-25T20:17:39Z"]},{"key":"dc:date.issued","label":"Date","values":["2023-12"]},{"key":"dc:type","label":"Dc Type","values":["thesis"]},{"key":"thesis:degree_discipline","label":"Discipline","values":["Computer science","Agriculture"]},{"key":"thesis:degree_level","label":"Degree Level","values":["thesis"]},{"key":"thesis:degree_name","label":"Degree Name","values":["Master of Science"]},{"key":"thesis:institution_name","label":"Thesis Institution Name","values":["Iowa State University"]}]},{"id":"language_rights","label":"Language and Rights","entries":[{"key":"dc:language.iso","label":"Language (ISO)","values":["en"]}]},{"id":"identifiers","label":"Identifiers","entries":[{"key":"dc:identifier.doi","label":"DOI","values":["https://doi.org/10.31274/td-20240617-205"]},{"key":"dc:identifier.uri","label":"Identifier URI","values":["https://dr.lib.iastate.edu/handle/20.500.12876/nrQBl5Az"]}]},{"id":"additional","label":"Additional Metadata","entries":[{"key":"dc:description.abstract","label":"Abstract","values":["Object detection faces a formidable challenge in detecting small objects due to their diminutive size and sparse pixel representation. This difficulty is exacerbated by the common lack of resolution in images, making the task of distinguishing small objects complex. Annotating data for training datasets also demands substantial manual effort and time, particularly when dealing with large numbers of small insects found in the field. Moreover, occlusion further complicates accurate detection, as small objects are prone to being obscured by other elements in the scene. To address these challenges, sophisticated techniques are needed, specifically tailored for small object detection within the broader field of object detection. In the context of insect detection, the task carries significant importance as certain insects pose a threat to agriculture and food security. Detecting these small pests in a timely manner is crucial to safeguarding food production and maintaining a stable agricultural economy. However, the insect detection task comes with various challenges, including the time-consuming annotation process for large datasets and difficulties in identifying small insects in low-resolution images or from a distance. Variations in insect life stages and similarities in shape and colors among different insect classes further add to the complexity. To overcome these challenges, this thesis explores and proposes two effective solutions: hierarchical transfer learning and slicing-aided hyper inference. Hierarchical transfer learning extends the concept of transfer learning by incorporating multiple steps of knowledge transfer from a large dataset to a target dataset. It leverages intermediate in-domain datasets to adapt the model progressively. On the other hand, slicing-aided hyper inference divides the original image into overlapping patches, enabling independent object detection on each patch and subsequent merging for accurate and comprehensive object detection outputs. The effectiveness of these proposed solutions was thoroughly evaluated through a comprehensive analysis, considering metrics such as mean Average Precision, Recall, and Precision. The evaluation focused on the accuracy of insect detection, especially in challenging scenarios. The thesis aims to contribute to more effective pest control measures, thereby safeguarding crops and supporting sustainable agricultural practices."]},{"key":"dc:format.mimetype","label":"Dc Format Mimetype","values":["PDF"]},{"key":"dc:title","label":"Title","values":["Hierarchical transfer learning for small object detection"]}]}],"canonical_facts":{"dc:contributor.advisor":["Tian, Jin","Sarkar, Soumik","Singh, Arti"],"dc:contributor.department":["Department of Computer Science"],"dc:creator":["Fotouhi, Fateme"],"dc:date.accessioned":["2024-01-25T20:17:39Z"],"dc:date.available":["2024-01-25T20:17:39Z"],"dc:date.issued":["2023-12"],"dc:description.abstract":["Object detection faces a formidable challenge in detecting small objects due to their diminutive size and sparse pixel representation. This difficulty is exacerbated by the common lack of resolution in images, making the task of distinguishing small objects complex. Annotating data for training datasets also demands substantial manual effort and time, particularly when dealing with large numbers of small insects found in the field. Moreover, occlusion further complicates accurate detection, as small objects are prone to being obscured by other elements in the scene. To address these challenges, sophisticated techniques are needed, specifically tailored for small object detection within the broader field of object detection. In the context of insect detection, the task carries significant importance as certain insects pose a threat to agriculture and food security. Detecting these small pests in a timely manner is crucial to safeguarding food production and maintaining a stable agricultural economy. However, the insect detection task comes with various challenges, including the time-consuming annotation process for large datasets and difficulties in identifying small insects in low-resolution images or from a distance. Variations in insect life stages and similarities in shape and colors among different insect classes further add to the complexity. To overcome these challenges, this thesis explores and proposes two effective solutions: hierarchical transfer learning and slicing-aided hyper inference. Hierarchical transfer learning extends the concept of transfer learning by incorporating multiple steps of knowledge transfer from a large dataset to a target dataset. It leverages intermediate in-domain datasets to adapt the model progressively. On the other hand, slicing-aided hyper inference divides the original image into overlapping patches, enabling independent object detection on each patch and subsequent merging for accurate and comprehensive object detection outputs. The effectiveness of these proposed solutions was thoroughly evaluated through a comprehensive analysis, considering metrics such as mean Average Precision, Recall, and Precision. The evaluation focused on the accuracy of insect detection, especially in challenging scenarios. The thesis aims to contribute to more effective pest control measures, thereby safeguarding crops and supporting sustainable agricultural practices."],"dc:format.mimetype":["PDF"],"dc:identifier.doi":["https://doi.org/10.31274/td-20240617-205"],"dc:identifier.uri":["https://dr.lib.iastate.edu/handle/20.500.12876/nrQBl5Az"],"dc:language.iso":["en"],"dc:title":["Hierarchical transfer learning for small object detection"],"dc:type":["thesis"],"thesis:degree_discipline":["Computer science","Agriculture"],"thesis:degree_level":["thesis"],"thesis:degree_name":["Master of Science"],"thesis:institution_name":["Iowa State University"]},"updated_at":"2026-07-24T02:37:49Z"}