Edge Kernel Filtered VGG Network based Fish Species Classification

Deep learning-based fish species categorization is a major development in the fields of ecological monitoring and aquatic biology. Conventional techniques for classifying species frequently depend on time-consuming, error-prone manual observation and professional expertise. In order to streamline the recognition and classification of fish species from underwater images, this study uses Deep Learning (DL) technology. With the view of technological advancements in engineering domain, this paper proposes Edge Kernel Filtered VGG16 (EKF-VGG16) that classifies the four classes of fish species more effectively. The Fish Species Dataset that includes 4000 fish images that was used for execution. The EKF-VGG16 model starts by organizes the fish images based on the four fish species type as coris, Boops, Solea and Anthais. During pre-processing, the organized fish images are subjected to labeled images, The labeled images are converted to gray scale image which is then processed with data augmentation to form 84,000 images. The augmented fish images are processed with edge filtering method applied with kernel to form Edge Kernel Filtered (EKF) fish images. The EKF fish images are with the existing CNN models and proposed EKF-VGG16 to assess the performance. With a high accuracy of 99.4%, the implementation shows that the suggested EKF-VGG16 outperforms in classifying the fish species type.