Dried fish variant classification using deep learning techniques

Abstract

This study describes a deep learning-based approach for digital classification of dried fish variants, which addresses an essential need in the food industry. With the increasing need for effective and precise methods of classifying dried fish variants, traditional evaluation methods have proven useless, which requires the study of advanced technological solutions. The proposed methodology includes data collection, labeling, image processing, model selection, training, evaluation, and testing. Local markets provided a diverse dataset with images of seven distinct dried fish variants, including Baspata Shutki, Bhangon, Chanda, Chapa, Chingri, Loitta, and Mola. Several deep learning architectures, including VGG16, InceptionV3, DenseNet201, and MobileNetV2, were tested for classification performance. The results show that convolutional neural networks (CNNs), especially DenseNet201, achieved high accuracy in classifying dried fish variants. Moral issues related to data privacy, fairness, and transparency were carefully addressed throughout the study. Further research should look into additional data sources, such as color imaging, transfer learning techniques, evaluating generalization capabilities across different contexts, including domain-specific information, and conducting ongoing research to monitor system effectiveness. Overall, this study advances automated food classification technologies and shows the transformative potential of deep learning in the food industry, allowing for greater efficiency, accuracy, and sustainability in dried fish variant classification processes.