{"id":{"repo_id":"msu","oai_identifier":"oai:d.lib.msu.edu:etd_53074"},"canonical_url":"https://search.ndltd.org/etd/msu/oai:d.lib.msu.edu:etd_53074","repository":{"repo_id":"msu","name":"Michigan State University","base_url":"https://d.lib.msu.edu/oai"},"display":{"title":"IMAGE-BASED FEED CHARACTERIZATION : FORAGE TYPE CLASSIFICATION AND MOISTURE CONTENT ESTIMATION USING DOMAIN-SPECIFIC SELF-SUPERVISED LEARNING","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":["Anuoluwapo, Akintan Oreofeoluwa"],"institution":null,"degree_name":null,"degree_level":null,"degree_discipline":null,"degree_department":null,"school":null,"contributors":["Uyeh, Daniel D","Morris, Daniel","Bradford, Barry"],"advisors":[],"committee_chairs":[],"committee_members":[],"year":2026,"date_issued":"2026","date_published":"2026","updated_at":"2026-11-18T18:59:37Z","subjects":["Agricultural engineering"],"languages":["eng"],"rights":["In Copyright"],"rights_urls":[],"identifier_entries":[{"key":"dc:identifier","label":"Identifier","values":["etd:53074","local:Anuoluwapo_grad.msu_0128N_22084"],"render_values":[{"text":"etd:53074","href":null,"code":true},{"text":"local:Anuoluwapo_grad.msu_0128N_22084","href":null,"code":true}]}],"person_identifier_entries":[]},"links":{"outbound_url":"https://doi.org/doi:10.25335/70ba-k976","outbound_label":"DOI","outbound_source":"dc:identifier","record_url":"https://doi.org/doi:10.25335/70ba-k976","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_53074","prefix":"oai_dc"},"metadata_groups":[{"id":"people","label":"People","entries":[{"key":"dc:contributor","label":"Contributor","values":["Uyeh, Daniel D","Morris, Daniel","Bradford, Barry"]},{"key":"dc:creator","label":"Author","values":["Anuoluwapo, Akintan Oreofeoluwa"]}]},{"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"]}]},{"id":"language_rights","label":"Language and Rights","entries":[{"key":"dc:language","label":"Dc Language","values":["eng"]},{"key":"dc:rights","label":"Dc Rights","values":["In Copyright"]}]},{"id":"identifiers","label":"Identifiers","entries":[{"key":"dc:identifier","label":"Identifier","values":["etd:53074","local:Anuoluwapo_grad.msu_0128N_22084","https://doi.org/doi:10.25335/70ba-k976"]}]},{"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 estimation of forage moisture content is critical for optimizing dry matter intake, feed quality, and storage stability in ruminant production systems. Conventional laboratory-based methods, while precise, are time-consuming and impractical for real-time, on-farm decision-making. This study presents a vision-based framework for forage characterization using RGB imaging and domain-specific self-supervised learning. A Latent-Euclidean Joint Embedding Predictive Architecture (LeJEPA) was employed to learn feature representations from 5,559 unlabeled forage images, which were subsequently evaluated on two downstream tasks: forage classification and moisture content estimation. Across backbone architectures, model performance varies depending on task and representation quality. For classification, ImageNet-pretrained models achieved the highest accuracy, reaching up to 99.25% in the full-data setting. In contrast, LeJEPA-pretrained models demonstrated competitive performance, particularly in data-limited regimes, with accuracy up to 96.27% (10-shot) and 92.54% (full dataset) depending on the architecture. For moisture estimation, LeJEPA-pretrained models showed strong predictive capability. The best-performing model, a custom convolutional neural network (CNN), achieved an R^2of 0.91, mean absolute error (MAE) of 4.53%, and root mean squared error (RMSE) of 6.08%. Other architectures, including EfficientNet-B0 and MobileNetV2, achieved R^2values of 0.77 and 0.74, respectively, but with higher error metrics. Notably, the custom CNN consistently outperformed deeper and more complex architectures across both classification and regression tasks. This performance advantage is attributed to the alignment between model capacity and dataset size. The custom CNN, with a moderate number of parameters, was better suited to the relatively limited labeled dataset, avoiding overfitting while still capturing domain-relevant features. In contrast, larger pretrained architectures exhibited diminished returns due to their higher parameter complexity relative to the available data. Overall, the results demonstrate that domain-specific self-supervised learning supports feature representation for forage analysis, particularly for continuous prediction tasks such as moisture estimation. Furthermore, the findings highlight the importance of matching model complexity to dataset scale, with appropriately sized architectures offering superior performance in data-constrained agricultural applications.","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 (70 p.) : ill."]},{"key":"dc:title","label":"Title","values":["IMAGE-BASED FEED CHARACTERIZATION : FORAGE TYPE CLASSIFICATION AND MOISTURE CONTENT ESTIMATION USING DOMAIN-SPECIFIC SELF-SUPERVISED LEARNING"]}]}],"canonical_facts":{"dc:contributor":["Uyeh, Daniel D","Morris, Daniel","Bradford, Barry"],"dc:creator":["Anuoluwapo, Akintan Oreofeoluwa"],"dc:date":["2026"],"dc:description":["Thesis (M.S.)--Michigan State University. Biosystems Engineering - Master of Science, 2026","Accurate estimation of forage moisture content is critical for optimizing dry matter intake, feed quality, and storage stability in ruminant production systems. Conventional laboratory-based methods, while precise, are time-consuming and impractical for real-time, on-farm decision-making. This study presents a vision-based framework for forage characterization using RGB imaging and domain-specific self-supervised learning. A Latent-Euclidean Joint Embedding Predictive Architecture (LeJEPA) was employed to learn feature representations from 5,559 unlabeled forage images, which were subsequently evaluated on two downstream tasks: forage classification and moisture content estimation. Across backbone architectures, model performance varies depending on task and representation quality. For classification, ImageNet-pretrained models achieved the highest accuracy, reaching up to 99.25% in the full-data setting. In contrast, LeJEPA-pretrained models demonstrated competitive performance, particularly in data-limited regimes, with accuracy up to 96.27% (10-shot) and 92.54% (full dataset) depending on the architecture. For moisture estimation, LeJEPA-pretrained models showed strong predictive capability. The best-performing model, a custom convolutional neural network (CNN), achieved an R^2of 0.91, mean absolute error (MAE) of 4.53%, and root mean squared error (RMSE) of 6.08%. Other architectures, including EfficientNet-B0 and MobileNetV2, achieved R^2values of 0.77 and 0.74, respectively, but with higher error metrics. Notably, the custom CNN consistently outperformed deeper and more complex architectures across both classification and regression tasks. This performance advantage is attributed to the alignment between model capacity and dataset size. The custom CNN, with a moderate number of parameters, was better suited to the relatively limited labeled dataset, avoiding overfitting while still capturing domain-relevant features. In contrast, larger pretrained architectures exhibited diminished returns due to their higher parameter complexity relative to the available data. Overall, the results demonstrate that domain-specific self-supervised learning supports feature representation for forage analysis, particularly for continuous prediction tasks such as moisture estimation. Furthermore, the findings highlight the importance of matching model complexity to dataset scale, with appropriately sized architectures offering superior performance in data-constrained agricultural applications.","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 (70 p.) : ill."],"dc:identifier":["etd:53074","local:Anuoluwapo_grad.msu_0128N_22084","https://doi.org/doi:10.25335/70ba-k976"],"dc:language":["eng"],"dc:relation":["Electronic Theses & Dissertations"],"dc:rights":["In Copyright"],"dc:subject":["Agricultural engineering"],"dc:title":["IMAGE-BASED FEED CHARACTERIZATION : FORAGE TYPE CLASSIFICATION AND MOISTURE CONTENT ESTIMATION USING DOMAIN-SPECIFIC SELF-SUPERVISED LEARNING"],"dc:type":["text","Electronic dissertations"]},"updated_at":"2026-11-18T18:59:37Z"}