Research On The Visual Attribute Recognition Method Of Factory Farmed Shrimp

In the process of aquaculture,visual observation and counting of shrimp in aquaculture pond is an important method to estimate the health status and density of object body.At present,aquaculture factories still use visual observation after fishing,which is not only inefficient,but also poor in objective consistency.Computer vision technology has developed rapidly in recent years,Based on this,a non-contact and intelligent method of monitoring and counting shrimp health status based on computer vision is proposed.Combined with the image resources of prawn real-time image acquisition system in culture pond,the application of object detection and object tracking algorithm for its visual attributes is studied.The specific work is summarized as follows:1)Noise reduction method of prawn image in culture pond: Aiming at the problem of poor quality of prawn image collected in actual culture environment,the noise reduction method is applied to prawn data set to provide high quality image for prawn detection.The experimental results show that the image noise reduction proposed in this paper can improve the detection performance of prawn.2)Research on health detection algorithm of underwater prawn based on improved YOLOv3.Aiming at the problem that the object individuals of shrimp are small and of different sizes.The loss function of object location in YOLOv3 algorithm is replaced by Loss CIOU,and CBAM attention mechanism and improved RFB receptive field module are added to the backbone network structure,and multi-scale feature fusion is carried out.Experimental results show that the improved algorithm increases the m AP value by 1.9%and the FPS value by 3.2%.3)Shrimp multi-object tracking and detection algorithm research.First use YOLOv5 algorithm to detect prawns in video,and then use DeepSORT algorithm to realize realtime tracking and counting of prawns.Using high-precision object detection algorithm to make up for the problem of low detection accuracy of DeepSORT algorithm.Figure 28;Table 7;Reference 58...