<![CDATA[Giant freshwater prawn (Macrobrachium rosenbergii) is a high-value freshwater fishery commodity. However, consumption eligibility assessment in the field often still relies on manual visual inspection, which can produce subjective results. This study develops an image classification system for eligible and non-eligible prawns using the Inception V3 architecture. The dataset consists of 7,000 images with two balanced classes, divided into 4,200 training images, 1,400 validation images, and 1,400 testing images. Data labeling was conducted based on visual and physical indicators, including color and texture, according to consumption eligibility criteria. The model was evaluated using SGD, Adam, and RMSprop optimizers, learning rates of 0.001, 0.0001, and 0.00001, and dropout rates of 0%, 20%, and 30%. Evaluation used accuracy, loss, confusion matrix, error analysis, and cost-benefit analysis. The best result was achieved by Inception V3 with 20% dropout, SGD optimizer, learning rate of 0.001, and batch size of 16. This configuration obtained a test accuracy of 0.9993 or 99.93%, with a Vi value of 0.9893. The trained model was implemented in a React Native-based Android application prototype for rapid field classification.]]>
Copyrights © 2026
