Dense Fish Population Counting Detection Based on Improved YOLOv7
[Objective]To improve the accuracy of fish school detection in complex environments such as turbid water bodies and high-density clustering of fish school.[Method]Dense fish population count detection method by an improved YOLOv7 is proposed,based on Vision Transformer with Bi-Level Routing Attention(BiFormer)and the Normalized Wasserstein Distance(NWD)loss function.On the basis of retaining fine-grained features,the model's ability to learn multi-scale features was improved,and the sensitivity of the model to the location deviation of small targets in fuzzy images was reduced,and the ability to identify fish in turbid waters was strengthened.In order to evaluate the effectiveness of the proposed model,comparative experiments with other network models were carried out on the dataset of the pond-cultured Takifugu rubripes.[Result]Comprehensive experimental results demonstrate that the precision and recall rate of the proposed method on the Takifugu rubripes dataset reach 98.05%and 97.69%respectively,and the average precision reaches 99.1%,which are 2.46%,3.73%and 2.62%higher than those of YOLOv7.The proposed model is also compared with current underwater target detection models with high detection accuracy.The average precision of the proposed model is increased by 4.25%on average.[Conclusion]The accurate detection of fish in turbid waters within real-world aquaculture environments is pivotal in guiding industrial aquaculture production and management with greater scientific precision.This advancement not only enhances aquaculture efficiency but also minimizes resource waste,thereby promoting sustainable aquaculture development.
aquaculturedetection of fishdeep learningYOLOv7BiFormerNWD
国家科技期刊平台
NSTL
万方数据
