基于改进自适应算法的小麦病虫害检测鲁棒性研究  

Research on the Robustness of Wheat Disease and Pest Detection Based on Improved Adaptive Algorithm

作　　者：杨绍清 李冰[1] YANG Shaoqing;LI Bing(Henan Polytechnic,College of Modern Information Technology,Zhengzhou 450046,China)

机构地区：[1]河南职业技术学院现代信息技术学院,河南郑州450046

出　　处：《传感器世界》2025年第1期15-21,共7页Sensor World

基　　金：河南省科技厅科技攻关(No.242102111190);河南省高等学校重点科研项目(No.24B520017)

摘　　要：小麦生长中,病虫害特征受环境、作物状态影响导致病虫害特征出现差异,进而加大检测难度。文章提出基于改进自适应算法的小麦病虫害检测鲁棒性研究,采用CDCNNv2模型学习小麦图像特征,捕捉病虫害在不同光照、时间下的细微变化,通过异常检测技术评估病虫害特征显著度,初步识别其类型和严重程度,引入自适应算法,计算平均检测精度,动态调整检测阈值和参数,结合收敛曲线和错误覆盖率确定最佳检测频次,实现检测效率最大化。实验结果显示,该方法在强光和弱光环境下均能保持高检测精度和低错误率,抗干扰性指数达0.9以上,证明了该方法的卓越性能和鲁棒性。In the growth of wheat,the characteristics of pests and diseases are affected by the environment and crop status,leading to differences in pest and disease characteristics and increasing the difficulty of detection.A study on the robustness of wheat pest and disease detection based on improved adaptive algorithms is proposed,using Convolutional Deep Belief Network Version 2(CDCNNv2)to learn wheat image features and capture subtle changes in pests and diseases under different lighting and time conditions,evaluate the significance of pest and disease characteristics through anomaly detection technology,and preliminarily identify their types and severity,introduce adaptive algorithms to calculate the average detection accuracy and dynamically adjust the detection threshold and parameters,determine the optimal detection frequency by combining the convergence curve and error coverage to maximize detection efficiency.The experimental results show that the proposed method can maintain high detection accuracy and low error rate in both strong and weak light environments,with an anti-interference index of over 0.9,the result proves the excellent performance and robustness of the proposed method.

关 键 词：小麦病虫害 异常检测 自适应算法 卷积深度神经网络模型 检测鲁棒性 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术]