CFSS-YOLO: A Detection Method for Cotton Top Bud in Real Farmland

Accurate identification of the cotton top bud is a prerequisite for automated cotton topping. However, the detection of the cotton top bud is low due to the small target size and a similar background. Therefore, a method named CFSS-YOLO was proposed to detect the cotton top bud based on YOLOv12 with an attention mechanism. Firstly, a Convolutional Block Attention Module (CBAM) was introduced into the neck structure of YOLOv12 to suppress background interference and improve target recognition accuracy. Secondly, a new loss function, FSSLoss, was designed where the Shape-IoU (Intersection over Union) optimized by Focaler-IoU was used for the part of localization loss, and Slideloss was integrated to improve the classification loss. The improvement of the loss function aimed to balance the relationship between classification loss and localization loss and accelerate the convergence speed of the model. The experimental results show that the precision, recall and mAP50 of the proposed CFSS-YOLO are 87.6%, 75.3% and 84.8%, respectively. The detection performance of the proposed method is superior to mainstream object detection models such as YOLOv12s, YOLOv5s, SSD, RT-DETR, and DEIM-R18 in outer farmland. These results demonstrate that the proposed CFSS-YOLO has high potential for application and promotion in the cotton top bud recognition task.