Fish Species Recognition using Deep Learning Techniques

Indian seashores and rivers have a major contribution in global fish production i.e., 5.43% which makes our country 2nd largest fish producer in the world. A local fisherman knows the commonly available fish’s structure and hence can easily identify the required species, but those fishes which are not so abundant and not known to fishermen can be easily identified with the help of deep learning technology. It is also important for the Marine Fishery Department to keep track of the population of fish species found in the regions and also maintain the count of such fish species which are endangered. Also, some fishes are harmful to humans or poisonous to consume and hence fishermen must know how to exactly distinguish edible and harmless fishes from the former ones. Thus, as a result, in this research, a cutting-edge smartphone app has been created for the identification of fish species that are found in India’s lagoons, coastal areas, including estuaries, and outer reef zones. The paper involves a detailed explanation of the methodologies for the Fish Species recognition. The goal of this research is to create a prototype system called Specifish that employs image processing techniques and deep learning frameworks to identify the species of saltwater fish that can be found in India. Dataset preparation is done using web scraping technique where data is scraped from fishbase website and labelled accordingly. The images are converted to the grayscale format. Image Augmentation is performed for sufficient availability of training data. In Automatic image classification and species recognition, we exploited different deep learning frameworks such as transfer learning techniques using models viz, MobileNet, ResNet, Inception. The experimental study shows that the outcome of our suggested model is promising.