Crop Recognition Based on Improved YOLOv9

• DOI: 10.62517/jike.202504110
• Publication Date: 2025-05-27T11:30:16Z
• Authors: Li Xiang; Jiao Xiangjian; Jia Maotang

To overcome challenges associated with crop species identification in unmanned aerial vehicle (UAV) imagery—specifically missed detections of small-sized targets, disturbances caused by intricate environmental elements, and suboptimal computational efficiency of detection models—this study introduces a streamlined detection framework that integrates YOLOv9 with Convolutional Block Attention Module (CBAM) and Spatial Channel Reconstruction Convolution (SCConv). Initial modifications involve embedding CBAM within the YOLOv9 backbone architecture. This enhancement exploits the collaborative functioning of channel-wise and spatial attention mechanisms to amplify the model's sensitivity toward critical localized crop characteristics, such as panicle morphological structures and foliar texture patterns, while concurrently mitigating noise interference from soil surfaces and weed vegetation. Subsequent improvements involve substituting conventional convolutional layers with Spatial Channel Reconstruction Convolution. This substitution capitalizes on the module's adaptive feature reorganization capabilities to achieve parameter reduction without compromising feature representation capacity, thereby substantially enhancing operational efficiency during edge device deployment. Empirical evaluations conducted on a proprietary dataset comprising 1,000 UAV-captured images representing seven distinct crop types demonstrate that the optimized model attains a mean Average Precision (mAP) of 94.5% when applying a 50% Intersection over Union threshold—a performance gain of 2.2 percentage points compared to the baseline YOLOv9 architecture. These results confirm the system's capability to meet real-time recognition demands within complex agricultural landscapes. The proposed methodology presents a cost-effective, high-accuracy monitoring solution for UAV-based precision agriculture applications, holding significant practical value for advancing intelligent management systems in modern agricultural practices.