Fish Classification Based on Underwater Image Interpolation and Back-Propagation Neural Network

The characteristics of the underwater environment affect the quality of underwater images. The low image resolution is one of major problems in the identification of fish species during monitoring of underwater ecosystems. Thus, the image only provides limited features, which affect the performance of classification methods. To the best knowledge of the authors, some prior studies merely focused on determining identification methods and often ignore the quality of the original data. To solve this research problem, this paper presents an image enhancement model applied to the process of fish species identification using the backpropagation neural network. This model is developed by choosing an appropriate interpolation method and an appropriate configuration of backpropagation neural network method until obtaining the best accuracy. The proposed method produced a new image with larger resolution resulted in the improvement of information that was contained in the image. Compared with traditional methods, our algorithm obtained higher accuracy in identifying the fish species as many as 90.24%. Therefore, the proposed method has the potential to support automatic fish identification system based on computer vision techniques.