Position prediction of the oyster adductor muscle based on YOLOv3 algorithm

Abstract

Oyster is one of the largest cultured shellfish in the world, though it remains a challenge to shuck oysters automatically by mechanical systems, which has attracted interests of research for a long time. We design a low-cost high-temperature steam beam to heat the adductor muscle attachment area with high precision to shuck the oysters. This approach, compared to the overall heating processes, causes much less damage to the quality and physiological structure of the oysters. The key issue of our method lies in locating the adductor muscle outside of the shells as there is no obvious feature of judgment due to the irregular shapes and variant sizes of the oysters. To this end, we proposed a deep learning method for predicting the position of the adductor muscle based on the YOLOv3 algorithm. In this paper, we establish an image dataset containing 520 oyster pictures, 120 of which are labeled pictures. These images are trained in the deployment environment of GTX 1060. Experiments show that the accuracy of the model is up to 99.5%, the prediction accuracy of the adductor muscle position reaches 79.17%, and the average time to detect one single image is around 0.03s.