With the introduction and deepening of the concept of sustainable aquaculture, the traditional aquaculture practices are gradually being supplemented and even replaced by advanced systems based on new technology. We here present an automated diagnostic method for diagnosis of cryptocaryoniasis in industrial marine aquaculture. It is based on computer image recognition technology targeting the typical clinical disease sign, white skin spots. A total of 800 images of healthy (400) and Cryptocaryon irritans infected (400) large yellow croaker (Larimichthys crocea) were obtained by cameras, and each type of image was enhanced to 1000. Based on the algorithm YOLOv3, the weights of the trained model yolov3.pt. were used to perform transfer learning on the enhanced image data set to establish the diagnosis model YOLOv3 of cryptocaryoniasis of L. crocea. Then, a visual real-time monitoring system for cryptocaryoniasis was developed. The results show that transfer learning could be well-applied to the training of the cryptocaryoniasis detection model. The accuracy of the final model was about 2% higher than that of the source model (average accuracy of YOLOv3 was 92%, recognition speed 36 frames/s). The algorithm YOLOv3 allows effective discrimination and recognition of cryptocaryoniasis in large yellow croaker. The visual real-time monitoring system allows the automatic and accurate diagnosis of cryptocaryoniasis in L. crocea aquaculture. The results illustrate the applicability of artificial intelligence for reduction of manpower expenditure in diagnostic work, shortening of detection time and elevation of accuracy and timeliness of problem recognition.
Read full abstract