基于YOLO的低成本且轻量化的秧苗计数研究OA北大核心CSTPCD

It is required to count the seedling in agricultural production to obtain assessment information such as seed quality and planting density.However,there is a high degree of manual dependence and poor hardware and software equipment in small-scale batch breeding environments.In view of the above,a low-cost and lightweight seedling counting method is proposed by taking Labiatae as an example.A fast construction method for the dataset is proposed to reduce the cost of the dataset.By improving the feature fusion part of the neck and streamlining the head of the YOLO model,the lightweight and efficiency of the model is ensured.In addition,an improved channel attention method is added to eliminate the excessive missed detection rate caused by the low resolution of the image.So far,the seedling counting model Seedet is constructed.The experimental results show that the Seedet's parameters and computational burden are reduced by 73.45%and 47.59%,respectively,and its detection accuracy and detection speed are improved by 3.987%and 70.09%,respectively,in comparison with those of the YOLOv5s.The proposed counting method is more suitable for seedling counting in low-cost scenarios and can promote the application of deep learning in agricultural production.