Abalone counting based on improved Faster R-CNN

Maintaining a reasonable culture density is a key element of abalone culture. In the early stage of aquaculture, when the individual size of abalone is small and the distribution is dense, the manual counting is slow and inaccurate. In this paper, based on the improved Faster R-CNN algorithm, the function of detecting and counting abalone individuals on farmed tiles was implemented. VGG16 is used as the backbone network for feature extraction, RPN network is used to generate suggestion box, and improved ROI pooling operation is adopted to make the algorithm more suitable for small-size abalone detection. Experimental results show that the proposed algorithm can detect abalone individuals in covered and shaded parts well.