Sea Turtles Species Classification With a Modified ResNet-50 Deep Learning Model

Sea turtles often undergo classification into distinct groups based on their species for conservation purposes. By classifying sea turtles in this way, conservationists can prioritize efforts and allocate resources effectively to protect these endangered and vulnerable species worldwide. Identifying sea turtle species is a major challenge when human visual inspection of the shells is required to determine the differences between species, which can lead to inaccuracies and errors in classification. Furthermore, such methods may disturb sea turtles in their natural environment. In this paper, we use images of sea turtle shells (carapace) of 7 species and preprocess them using the Sobel operator to improve edge detection and feature detection in images with different levels of noise and detail and a Color Histogram to find the distinctive features from the color distribution in the image from the pixel intensity distribution in the image to be complete and suitable for learning by using ResNet-50, a deep convolution neural network architecture, to identify sea turtle species accurately. The experimental results of marine species classification achieved an accuracy with an average of 90.4%, proving that it is an effective and feasible tool for marine conservation applications. Monitor and protect endangered sea turtle populations without disturbing or harming their natural habitats.