Quantitative detection method of swimming activity of Litopenaeus vannamei based on improved YOLOv7-tiny

Shrimp is rich in a variety of trace elements and vitamins, and has a substantial nutritional value, it is also be widely recognized as an important ingredient in high-end, well-known cuisine. Among them, the cultural production of Litopenaeus vannamei accounts for about 85% of the total production of shrimp culture, which is an important economic aquaculture object. The active state of L. vannamei reacts its health condition and behavioral situation. Surveying and identifying the activity of L. vannamei is helpful in finding abnormal behavior in aquaculture, to give early warning and take remedial methods promptly, lessen economic losses in aquaculture, and improve the yield and efficiency of aquaculture. Nowadays in the L. vannamei pond aquaculure process, aquaculture personnel often need to monitor the active swimming state of the shrimp by manually pulling the feed tray, then analyzing the overall environment of the aquaculture pond and formulating effective aquaculture breeding strategies. However, due to the complexity of the pond underwater environment, artificial observation experience is limited, so the method of manually observing the active state of L. vannamei has a lot of problems, such as inefficiency limited scope of application, low accuracy, poor real-time performance, high labor intensity and other problems. In order to solve these problems, propose a visual detection method for the activity of L. vannamei based on an improved YOLOv7 tiny network detection model and multi-objective association based on Euclidean distance to quantitatively study the swimming activity status of shrimp. Based on the YOLOv7 tiny network model, the standard convolution was replaced by Conv convolution, and a VoVGSCSPC module was built to replace the original lightweight aggregation module (ELAN-L). The MPDIoU loss function was used instead of the CIoU loss function to reduce the model capacity and improve the model detection accuracy. The position of shrimp in the image was determined by the visual detection results of improved YOLOv7-tiny model and the multi-objective association method based on Euclidean distance, from which the shrimp's swimming displacement, speed and turn angle were calculated to quantify the shrimp's swimming activity status. After validation on the L. vannamei dataset, the results showed that the misdetection rate and omission rate of the improved YOLOv7-tiny model were reduced by 0.62% and 1.05%, respectively, compared with the YOLOv7-tiny model. The inference speed was improved by 17.07%, so the effectiveness of the improved model was verified. Quantitative analysis of the activity of shrimp showed that the more active shrimp corresponded to the higher the activity index value, which was consistent with the actual situation. The study showed that the proposed quantitative detection method could accurately and quickly obtain the swimming activity index, and could efficiently quantify the swimming activity state of L. vannamei on the feed tray, which was of great significance to grasp the health status of L. vannamei and improved the intelligent level of shrimp culture.