00000ntmaa22000002i 4500 UP-8027390931312009139 Buklod 20260212102326.0 m|||||o||d|||||||| cr ||||||||||| 260212s2025 xx a b ||| u eng UPVT-00020035100 UPTC rda eng LG 993.5 2025 C66 P45 Phillips, Charles Roy R. Jr. author. MarketNet multicolor hybrid CNN and ViT for multitask image classification & segmentation Charles Roy R. Phillips, Jr. ; John Paul T. Yusiong, adviser. Tacloban City Division of Natural Sciences and Mathematics, University of the Philippines Tacloban College 2025. xiii, 91 leaves illustrations, color 31 cm. text rdacontent unmediated rdacontent volume rdacarrier Undergraduate thesis (Bachelor of Science in Computer Science) -- University of the Philippines, Tacloban. Includes bibliographical references. Available to the general public-YES. Available only after consultation with author/adviser-NO. Available only for those bound by confidentiality agreement-NO. Monitoring grocery inventory is important for effective supply chain management and reduction of waste. Current grocery image classification models generally rely on RGB images but fail to capture the complex variations of grocery items. This paper aims to counter that limitation by introducing the novel deep learning model, known as MarketNet, which specifically aims at grocery image-based classification and segmentation tasks using multi-input channels. MarketNet uses a multi-input convolutional neural network (CNN) integrated with a Vision Transformer (VIT) to improve classification and segmentation accuracy The model incorporates multicolor input to learn richer and more discriminative features. The method was tested using a grocery image dataset in which the model classified items into predefined categories while simultaneously segmenting images for further localization. The methodology also includes preprocessing grocery images into multiple color channels followed by feature extraction by CNN. Features are further improved by the Vision Transformer to better accuracy and efficiency in predicting. It is shown experimentally that MarketNet does better in comparison to its baseline variants with the following metrics 98.87% Top-1 accuracy and 99.93% Top-5 accuracy for classification, and 94.54% mIoU, 96.73% accuracy, 96.81% precision, 97.54% recall, and 97.19% F1-Score far segmentation. This work contributes to the progress of grocery image analysis as a framework that can support improved inventory management, reduction in waste, and real-time product tracking within grocery settings. Inventory control Automation. Image processing Digital techniques. Yusiong, John Paul T. adviser. Thesis FI UP UPTAC UPTAC LG 993.5 2025 C66 P45 Thesis