Detecting Phenotypes of Seedling Soybean with Enhanced Detection Head and Attention Mechanism

Accurate detection of soybean phenotypes is crucial for growth stage monitoring and crop management. The detection model must cope well with soybean phenotypes with small sizes and complex shapes. This paper proposes a seedling soybean phenotype detection method based on the improved Faster R-CNN and the attention mechanism. We introduce the Convolutional Block Attention Module (CBAM) to the backbone to help the model focus on target object areas. Also, an enhanced fully convolutional (EFC) head was proposed in order to improve the detection accuracy of phenotypes with complex shapes. Instead of smooth L1 loss, using of CIoU loss helps the model convergence easier. The proposed method is tested on datasets collected in real agricultural fields. The experimental results show that the proposed method surpasses the performance of other SOTA object detection algorithms. The method achieves promising performance in seedling soybean phenotype detection, and is of benefit to crop monitoring and management.