改进RetinaNet的农作物病虫害检测模型研究

Research on Improving RetinaNet’s Crop Pest Detection Model

【目的】为快速准确检测密集遮挡环境下农作物病虫害情况,满足大规模、高效率的识别需求。研究提出了一种改进RetinaNet的农作物病虫害检测模型。【方法】结合DenseNet改造RetinaNet的特征提取网络,强化特征重用,提高深度网络模型对农作物害虫的识别率,准确、快速地识别出病虫害的种类。其非极大抑制算法采用了DIoU策略,有效降低了在目标密集的情况下农作物病虫害的漏检率,提高了检测精度。【结果】改进后的算法模型具有较高的检测精度和良好的实时性,对作物密集遮挡情况有较好的适应性,mmAP达到了49.8%。【结论】将模型应用于复杂的农作物生长环境中,能有效提升病虫害智能检测能力。

【Objective】 To quickly and accurately detect crop pests and diseases in dense shading environments to meet the needs of large-scale and high-efficiency identification. The study proposes an improved RetinaNet crop pest detection model. 【Method】 Combines with DenseNet, the RetinaNet feature extraction network was modified to enhance feature reuse, improve the recognition rate of the deep network model for crop pests, and identify the species of pests and diseases accurately and quickly. Its nongreat suppression algorithm adopts the DIoU strategy, which effectively reduces the leakage rate of crop pests in the case of dense targets and improves the detection accuracy. 【Result】 The experimental results show that the improved algorithm model has high detection accuracy and good real-time performance, and has good adaptability to the dense crop shading situation, and the mmAP reaches 49.8%.【Conclusion】 Applying the model to the complex crop growth environment can effectively improve the intelligent detection of pests and diseases.