{"id":{"repo_id":"msu","oai_identifier":"oai:d.lib.msu.edu:etd_53141"},"canonical_url":"https://search.ndltd.org/etd/msu/oai:d.lib.msu.edu:etd_53141","repository":{"repo_id":"msu","name":"Michigan State University","base_url":"https://d.lib.msu.edu/oai"},"display":{"title":"A Multimodal Dataset and Image-Based Framework for Livestock Feed Characterization","abstract":"Thesis (M.S.)--Michigan State University. Biosystems Engineering - Master of Science, 2026","abstract_html":"Thesis (M.S.)--Michigan State University. Biosystems Engineering - Master of Science, 2026","abstract_has_math":false,"creators":["Nwaneri, Ifeanyi Vincent"],"institution":null,"degree_name":null,"degree_level":null,"degree_discipline":null,"degree_department":null,"school":null,"contributors":["Uyeh, Daniel","Morris, Daniel","Bradford, Barry"],"advisors":[],"committee_chairs":[],"committee_members":[],"year":2026,"date_issued":"2026","date_published":"2026","updated_at":"2026-11-18T18:59:41Z","subjects":["Agricultural engineering","Computer science"],"languages":["eng"],"rights":["Attribution 4.0 International"],"rights_urls":[],"identifier_entries":[{"key":"dc:identifier","label":"Identifier","values":["etd:53141","local:Nwaneri_grad.msu_0128N_22156"],"render_values":[{"text":"etd:53141","href":null,"code":true},{"text":"local:Nwaneri_grad.msu_0128N_22156","href":null,"code":true}]}],"person_identifier_entries":[]},"links":{"outbound_url":"https://doi.org/doi:10.25335/hbkf-1539","outbound_label":"DOI","outbound_source":"dc:identifier","record_url":"https://doi.org/doi:10.25335/hbkf-1539","record_label":"Repository record","record_kind":"doi","record_source":"dc:identifier","download_url":null,"download_label":null,"download_source":null,"access_status":null,"embargo_until":null,"access_note":null},"source_record":{"url":"https://d.lib.msu.edu/oai?verb=GetRecord&metadataPrefix=oai_dc&identifier=oai%3Ad.lib.msu.edu%3Aetd_53141","prefix":"oai_dc"},"metadata_groups":[{"id":"people","label":"People","entries":[{"key":"dc:contributor","label":"Contributor","values":["Uyeh, Daniel","Morris, Daniel","Bradford, Barry"]},{"key":"dc:creator","label":"Author","values":["Nwaneri, Ifeanyi Vincent"]}]},{"id":"academic_context","label":"Academic Context","entries":[{"key":"dc:date","label":"Dc Date","values":["2026"]},{"key":"dc:relation","label":"Dc Relation","values":["Electronic Theses & Dissertations"]},{"key":"dc:type","label":"Dc Type","values":["text","Electronic dissertations"]}]},{"id":"subjects_keywords","label":"Subjects and Keywords","entries":[{"key":"dc:subject","label":"Dc Subject","values":["Agricultural engineering","Computer science"]}]},{"id":"language_rights","label":"Language and Rights","entries":[{"key":"dc:language","label":"Dc Language","values":["eng"]},{"key":"dc:rights","label":"Dc Rights","values":["Attribution 4.0 International"]}]},{"id":"identifiers","label":"Identifiers","entries":[{"key":"dc:identifier","label":"Identifier","values":["etd:53141","local:Nwaneri_grad.msu_0128N_22156","https://doi.org/doi:10.25335/hbkf-1539"]}]},{"id":"additional","label":"Additional Metadata","entries":[{"key":"dc:description","label":"Description","values":["Thesis (M.S.)--Michigan State University. Biosystems Engineering - Master of Science, 2026","Accurate and timely characterization of livestock feed is essential for effective ration formulation, optimizing animal performance, and sustainable livestock production. However, existing assessment methods, including laboratory-based chemical analysis, static feed composition tables, and advanced spectroscopic systems, are often slow, costly, or inaccessible under real-world production conditions, particularly in smallholder and resource-limited settings. This thesis addresses these limitations through two interconnected contributions that establish data standardization and component-aware image analysis as foundational elements of a scalable, interpretable framework for livestock feed evaluation. The first contribution is a standardized, multi-country multimodal livestock feed dataset comprising more than 1,100 samples collected across sixteen countries spanning diverse agroecological zones and production systems. Each sample is linked to high-resolution RGB images, four-band multispectral images, laboratory-verified nutritional composition, and structured field metadata capturing feed type, maturity stage, processing form, and geographic context. Harmonized sampling, imaging, and analytical protocols were implemented across all participating institutions to ensure consistency and cross-country comparability. Cross-laboratory validation between national laboratories and a centralized reference facility revealed systematic analytical differences across three countries, highlighting the need for standardized protocols in multi-institutional feed research. This dataset provides a robust, sample-level foundation for statistical analysis, machine learning model development, and benchmarking of image- and sensor-based feed assessment methods. The second contribution is an automated synthetic data generation pipeline and a U-Net++ semantic segmentation model for identifying leaf and stem components in heterogeneous chopped forage mixtures from RGB images. To address the lack of annotated training data, a compositing strategy was developed to generate pixel-level labeled synthetic images directly from isolated photographs of forage components, eliminating the need for manual annotation. A model trained exclusively on this synthetic data achieved a mean Intersection-over-Union (mIoU) of 0.68, a mean Dice coefficient of 0.80, and a pixel accuracy of 0.84 on synthetic validation images. When evaluated on forty manually annotated real-world forage images, the model obtained an mIoU of 0.65, a mean Dice coefficient of 0.76, and a pixel accuracy of 0.87, demonstrating effective generalization from synthetic training conditions to real, unstructured feed environments. The practical deployment of this model is illustrated by a prototype mobile application that supports the vision of accessible, on-farm feed assessment using widely available devices. Together, these contributions establish a pathway toward interpretable, image-based nutrient prediction in future work. Structural features derived from segmentation, such as leaf-to-stem ratio, combined with contextual metadata captured in the multimodal dataset, including maturity stage, harvest season, and species identity, provide the feature infrastructure needed to train predictive models for key nutritional outcomes such as crude protein, NDF, and ADF. This thesis, therefore, positions semantic segmentation not as an endpoint but as an interpretable intermediate step that bridges visual feed observation and quantitative nutrient estimation, contributing a practical and scalable framework for precision livestock nutrition across diverse global production environments.","Description based on online resource. Title from PDF t.p. (Michigan State University Fedora Repository, viewed ).","Includes bibliographical references."]},{"key":"dc:format","label":"Dc Format","values":["application/pdf","1 online resource (87 p.) : ill."]},{"key":"dc:title","label":"Title","values":["A Multimodal Dataset and Image-Based Framework for Livestock Feed Characterization"]}]}],"canonical_facts":{"dc:contributor":["Uyeh, Daniel","Morris, Daniel","Bradford, Barry"],"dc:creator":["Nwaneri, Ifeanyi Vincent"],"dc:date":["2026"],"dc:description":["Thesis (M.S.)--Michigan State University. Biosystems Engineering - Master of Science, 2026","Accurate and timely characterization of livestock feed is essential for effective ration formulation, optimizing animal performance, and sustainable livestock production. However, existing assessment methods, including laboratory-based chemical analysis, static feed composition tables, and advanced spectroscopic systems, are often slow, costly, or inaccessible under real-world production conditions, particularly in smallholder and resource-limited settings. This thesis addresses these limitations through two interconnected contributions that establish data standardization and component-aware image analysis as foundational elements of a scalable, interpretable framework for livestock feed evaluation. The first contribution is a standardized, multi-country multimodal livestock feed dataset comprising more than 1,100 samples collected across sixteen countries spanning diverse agroecological zones and production systems. Each sample is linked to high-resolution RGB images, four-band multispectral images, laboratory-verified nutritional composition, and structured field metadata capturing feed type, maturity stage, processing form, and geographic context. Harmonized sampling, imaging, and analytical protocols were implemented across all participating institutions to ensure consistency and cross-country comparability. Cross-laboratory validation between national laboratories and a centralized reference facility revealed systematic analytical differences across three countries, highlighting the need for standardized protocols in multi-institutional feed research. This dataset provides a robust, sample-level foundation for statistical analysis, machine learning model development, and benchmarking of image- and sensor-based feed assessment methods. The second contribution is an automated synthetic data generation pipeline and a U-Net++ semantic segmentation model for identifying leaf and stem components in heterogeneous chopped forage mixtures from RGB images. To address the lack of annotated training data, a compositing strategy was developed to generate pixel-level labeled synthetic images directly from isolated photographs of forage components, eliminating the need for manual annotation. A model trained exclusively on this synthetic data achieved a mean Intersection-over-Union (mIoU) of 0.68, a mean Dice coefficient of 0.80, and a pixel accuracy of 0.84 on synthetic validation images. When evaluated on forty manually annotated real-world forage images, the model obtained an mIoU of 0.65, a mean Dice coefficient of 0.76, and a pixel accuracy of 0.87, demonstrating effective generalization from synthetic training conditions to real, unstructured feed environments. The practical deployment of this model is illustrated by a prototype mobile application that supports the vision of accessible, on-farm feed assessment using widely available devices. Together, these contributions establish a pathway toward interpretable, image-based nutrient prediction in future work. Structural features derived from segmentation, such as leaf-to-stem ratio, combined with contextual metadata captured in the multimodal dataset, including maturity stage, harvest season, and species identity, provide the feature infrastructure needed to train predictive models for key nutritional outcomes such as crude protein, NDF, and ADF. This thesis, therefore, positions semantic segmentation not as an endpoint but as an interpretable intermediate step that bridges visual feed observation and quantitative nutrient estimation, contributing a practical and scalable framework for precision livestock nutrition across diverse global production environments.","Description based on online resource. Title from PDF t.p. (Michigan State University Fedora Repository, viewed ).","Includes bibliographical references."],"dc:format":["application/pdf","1 online resource (87 p.) : ill."],"dc:identifier":["etd:53141","local:Nwaneri_grad.msu_0128N_22156","https://doi.org/doi:10.25335/hbkf-1539"],"dc:language":["eng"],"dc:relation":["Electronic Theses & Dissertations"],"dc:rights":["Attribution 4.0 International"],"dc:subject":["Agricultural engineering","Computer science"],"dc:title":["A Multimodal Dataset and Image-Based Framework for Livestock Feed Characterization"],"dc:type":["text","Electronic dissertations"]},"updated_at":"2026-11-18T18:59:41Z"}