A Comparative Analysis of Mobilenetv3small, Mobilenetv3large, and Efficientnetb0 for Chicken Meat Freshness Classification

Chicken meat is a food item widely consumed by the public because it is a good source of protein and relatively easy to obtain. The freshness of chicken meat is an important aspect that must be considered before consumption; however, visual inspection can still be subjective and dependent on the user’s experience. Advances in image processing and deep learning offer opportunities to automate the identification of chicken meat freshness based on visual characteristics. This study aims to conduct a comparative analysis of three Convolutional Neural Network (CNN) architectures—MobileNetV3Small, MobileNetV3Large, and EfficientNetB0—for classifying chicken meat images into two classes: fresh and not fresh. The dataset consists of 276 chicken meat images, divided into 131 images in the “fresh” class and 145 images in the “not fresh” class. All models employ a transfer learning approach by utilizing pretrained weights, allowing the models to adapt to the relatively limited research dataset. Data processing steps included resizing images to 224×224 pixels, normalization, and augmentation of the training data. All three architectures were trained and evaluated using the same data configuration and metrics to allow for a more objective comparison of performance. The analysis is based on classification performance and the characteristics of each architecture. The research results are expected to highlight the differences in the capabilities of MobileNetV3Small, MobileNetV3Large, and EfficientNetB0 in recognizing the freshness of chicken meat, while also providing considerations for selecting a model that strikes a balance between accuracy.