Shrimpseed-Net: Counting of Shrimp Seed Using Deep Learning on Smartphones for Aquaculture

Abstract

Accurate shrimp seed counting is an important task in the aquaculture and stocking shrimp seeds processes. At present, most shrimp seed counting tasks are still performed manually, which is not only time-consuming and laborious but also has low accuracy, and this counting method cannot realize modern aquaculture. Therefore, to promote the modernization of aquaculture, a portable shrimp seed counting system is proposed in this paper. The system is based on Shrimpseed-Net, a modified CSRNet convolutional neural network, and can achieve higher accuracy in shrimp seed counting. The algorithm has been successfully implemented on a smartphone, enabling automatic counting of shrimp seeds within seconds by simply taking a picture of the shrimp seeds or selecting an image from the photo album to upload to the internal algorithm. Experimental results show that after only 50 iterations, the average absolute error of the algorithm is reduced to 17.28, with an accuracy rate of 95.53%. The system can accurately count shrimp seeds in real scenarios within a short period. The shrimp seed counting system studied in this paper can be widely used in the aquaculture industry to promote its development and enable intelligent aquaculture practices.