Detecting Fish in Images with Complex Background and Estimating Body Length using Convolutional Neural Network = Detecting Fish in Images with Complex Background and Estimating Body Length using Convolutional Neural Network

다국어 초록 (Multilingual Abstract)

Fish body length is considered as an important index for resource management. Many organizations put restrictions on the size of caught fish that can be retained. In conservation ecology, fish body length is also used as an indicator for determining the sexual maturity. Conventionally, fish body length was measured manually using rulers or tape measures. Manual methods are, however, time consuming, labor intensive, imprecise and subjective. This study proposed a method to automatically measure fish body length from images with complex background. In the approach, a convolutional neural network classifier was first developed to detect fish head, fish caudal, and color plate in an image. Pixel to distance ratio was then calculated using the known length (25cm) of the color plate. Next, fish body length was estimated as the distance between the fish head and caudal. The approach reached an accuracy of 95.53% for fish body length estimation.