Research on Cotton Seedling Detection Method Based on YOLOv8

In response to the deployment requirements of the cotton seedling detection model on edge devices, this paper proposes a lightweight detection model to accommodate the limited computational resources of edge environments. First, the FasterNet is used to replace the backbone network, reducing redundant computations and memory accesses, thereby enhancing the detection speed of the model. Secondly, the NAM attention mechanism is introduced to enhance the detection accuracy of small objects, strengthen feature extraction across different channel spaces, and simultaneously suppress the weight proportions of less important regions. Finally, the C2f in the Neck section is replaced with C2f-DCNv3 to flexibly adapt to different types of transformations when handling feature outputs of varying shapes, thereby enhancing the capability to extract features of different shapes. Experimental results indicate that the improved YOLOv8 achieves an average detection accuracy of 91.8% for cotton seedlings, which is an increase of 1.8% compared to the original model, while the floating point operations have decreased by 24.7<sup xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">%</sup>. The experiments demonstrate that our proposed FND-YOLOv8 significantly improves the real-time detection speed of cotton seedlings while enhancing the model's detection accuracy, providing a viable solution for future deployment on mobile devices.