RIASSUNTO
Detection and counting small objects using under-water robot draw an appealing attention because of its urgent demands in marine aquaculture. Because this challenge problem must be solved before the underwater robot can be used to catch seafood in practice instead of diver. This paper proposed a novel method using Faster R-CNN and kernelized correlation filter (KCF) tracking algorithm to detect seafood objects, such as sea cucumber, sea urchin, and scallop and so on in real time. Firstly, we trained an accurate and stable Faster R-CNN detector with VGG model using underwater image database, which is built by ourselves. Next, we recognized and tracked the seafood objects in order to fetch them using underwater robot vision in naturalistic ocean environment. The experimental results show the proposed method can recognized and catch seafood in real time using our integrated underwater robot.