Flounders Measurement System Using Obtained Optical Images by AUV

The observation of marine life by autonomous underwater vehicle and the marine life counting and measurement system are highly required for the fisheries resources management. In this research, we propose an automatic flounder, which is valuable as a food resource, length measurement system. The proposed system consists of the image enhancement process, selection region of candidate process, recognition process and measurement process. In the image enhancement process, the Retinex theory is employed to improve visibility of seafloor image, and the enhanced image used to select candidate region in the selection region of candidate process. Then, AlexNet, which is object recognition algorithm based on convolutional neural networks, employed for the flounder recognition, and the length of recognized flounder is estimated by measurement process. Flounders are inhabited on seafloor and the shape does not change significantly, by these properties, the length of flounders is estimated using a straight line from head to tail in the measurement process. For evaluation, 10 seafloor images, which are obtained by AUV, are used. The result of evaluation, the region selection error was 31%, recognition accuracy was 85% and the average of error of estimated flounder's length was 6 pixels.