Automatic measurement of fish from images using convolutional neural networks

The morphological characteristics of fish, such as body length and width, are frequently used in ichthyology and fisheries studies. Currently, the information collection from fish depends mainly on manual measurement, which is operationally complex and inefficient. In order to solve these problems, this work proposed a method to automate the collection of morphological characteristics of freshwater fish using the Mask R-CNN network, and techniques were used to calculate and find the three lengths of the fish. The results show that the proposed method manages to segment the images even when dealing with several categories of fish and handling images with complex backgrounds, achieving 82.21% with the IOU metric and 83.89% with the pixel accuracy metric. When measuring the morphological characteristics of the fish, the average relative error for total, furcal, and standard length were all below 10%. © The Author(s), under exclusive licence to Springer Science+Business Media, LLC, part of Springer Nature 2024.