A Small Object Detection Method for Cordyceps Based on Improved YOLOv8

Object detection technology holds immense poten-tial and is increasingly being applied across various domains. One such application is the use of object detection techniques for locating Cordyceps. However, detecting small objects like Cordyceps presents challenges due to their small size in images and susceptibility to environmental factors. To enhance the model's detection capability for small objects, We have adopted the data processing method of image cropping, to increase the relative size of small targets, thereby improving feature extraction efficiency, and the same cropping method is also employed during actual inference. Furthermore, in the model architecture, we introduced a small object detection layer that integrates multi-scale features. Additionally, we leveraged fully Omni-Dimensional Dynamic Convolution to adaptively learn the features of small objects. Experimental results demonstrate a significant performance boost, achieving a precision of 0.961 (mAP@50), which is comparable to YOLOv8s with triple the parameter count. Although the performance of the model en-hanced, further research and improvement are needed to address the limitations of a singular data processing approach, fixed image cropping sizes that don't adapt to varying object sizes, and increased inference time.