Research on counting method of half-smooth tongue sole fry based on machine vision

In response to the problems of large errors and low efficiency in the current stage of manual completion of fry counting of half-smooth tongue sole, this paper proposes a counting method based on machine vision technology, which consists of a counting algorithm combining skeleton endpoints and crosspoints and a compensation algorithm based on this counting algorithm. First, the fry images are acquired and preprocessed to obtain the binarized images, all the connected domain parameters are acquired and the corresponding connected domain images are constructed, and the fry skeleton images are obtained using the skeleton thinning algorithm. Subsequently, the defects of the endpoint counting method were analyzed. The crosspoint counting method was proposed, and new rules for determining endpoints, skeleton points and crosspoints were established. The two counting methods were compared and analyzed under different fry distribution situations. Then, a compensation algorithm based on crosspoint counting is proposed. The process of obtaining the average standard area of fry and how to set the scale factor reasonably are presented. Finally, the fry images were divided into three groups, A, B and C, according to the degree of overlap and counted using two methods. The results showed that the accuracy of endpoint counting method was 79.97%, 64.06% and 46.28%, and the accuracy of crosspoint counting method was 98.49%, 98.65% and 98.93%, respectively.