Instance Segmentation of Golden crucian carp based on Cascade Mask RCNN

Precision fishing and real-time monitoring are growing trends toward "intelligent" fishery. However, the fishery is still in a relatively backward artificial aquaculture due to the complex underwater environment and the intricate movement posture of fish. To realize the precision aquaculture of fishery, this paper took golden crucian carp for example selecting 10 golden crucian carp as experimental objects and formed 640 golden crucian carp instance segmentation data set through labelme manual labelling method. This paper proposes a method to solve fish segmentation and distinguish individuals. The solution optimises the cascade mask RCNN model, so that the model could effectively perform case segmentation of gold crucian carp. By comparing the predictive models and multiple case segmentation models, the former performs better. Among them, the mAP of Bbox of Cascade Mask RCNN reached 0.916, and the mAP of segmentation also reached 0.917, which can effectively complete the task of individual differentiation and body size estimation for Gold crucian carp. This study provides a data set and reference for body size estimation in fish farming.