Automatic Detection of Nephrops norvegicus Burrows in Underwater Images Using Deep Learning

Abstract

Autonomous Underwater Vehicles and Remotely Operated Vehicles equipped with HD cameras are used by the scientist to capture the underwater footages efficiently and accurately. The abundance of the Norway Lobster Nephrops norvegicus stock in the Gulf of Cadiz is assessed based on the identification and counting of the burrows where they live, using underwater videos. The Instituto Español de Oceanografía (IEO) conducts an annual standard underwater television survey (UWTV) to generate burrow density estimates of Nephrops within a defined area, with a coefficient of variation (CV) or relative standard error of less than 20%. Currently, the identification and counting of the Nephrops burrows are carried out manually by the experts. This is quite hectic and time consuming job. Computer Vision and Deep learning plays a vital role now a days in detection and classification of objects. The proposed system introduces a deep learning based automated way to identify and classify the Nephrops burrows. The proposed work is using current state of the art Faster RCNN models Inception v2 and MobileNet v2 for objects detection and classification. Tensorflow is used to evaluate the Inception and MobileNet performance with different numbers of training images. The average mean precision of Inception is more than 75% as compared to MobileNet which is 64%. The results show the comparison of Inception and MobileNet detections, as well as the calculation of True Positive and False Positive detections along with undetected burrows.