基于视觉识别的玉米病虫害检测与精准变量喷药系统研究

Maize Pest and Disease Detection and Precise Variable Spraying System Based on Visual Recognition

摘要:

针对传统无差别连续式喷药存在农药浪费、喷施低效的问题，以玉米为研究对象，设计一套基于视觉识别的病虫害检测及精准变量喷药系统。结合图像处理和机器视觉技术，对玉米田间病虫害自动、快速和准确识别，并根据识别的病虫害种类及严重程度，自动调整喷药剂量，实现精准农业管理。将自主设计的变量喷药系统集成并部署于计算机控制系统中，并对其检测性能进行验证，试验结果表明，相较于基准模型 YOLO v5s，改进后模型精确率（P）、召回率（R）、mAP值分别提升1.6、1.3、0.7个百分点，降低了病虫害误检，避免对非病虫害区域的误喷，同时减少漏检确保了病虫害区域得到及时有效处理，综合反映了系统在不同病虫害类别上的整体识别能力;对于玉米螟、黏虫、灰斑病、叶斑病和锈病，模型识别准确率稳定在60%以上，而对于红蜘蛛、蚜虫识别准确率则在40%以上。于田间进行喷药性能试验，并对雾滴沉积、雾滴漂移及省药率等关键指标进行测试与分析，结果表明，最低雾滴覆盖率为52%，最低平均沉积密度为71.3滴/cm2，均达到病虫害防治要求;省药率与地面流失率最低值分别为32.1%和22%，显著降低了农药总体消耗量和地面流失率。本文设计的玉米病虫害检测及精准变量喷药系统，显著提升了病虫害识别准确性，提高了农药利用率并降低环境污染，为病虫害防控提供科学高效的解决方案。

Abstract:

A set of pest and disease detection and precise variable spraying system, based on visual recognition, was designed for maize, addressing traditional issues of pesticide waste and spraying inefficiency. Utilizing image processing and machine vision, this system automatically and accurately identified pests and diseases in maize fields, adjusting spraying doses accordingly. It was then integrated into a computer control system, with its performance verified. The system surpassed the benchmark model YOLO v5s, improving P, R, and mAP by 1.6，1.3 and 0.7 percentage points，respectively. The high precision rate reduced false detection of pests and diseases to avoid false spraying of non-pest areas. The high recall rate reduces missed detection to ensure timely and effective treatment of pest and disease areas. The improvement of mAP value comprehensively reflected the overall identification ability of the system in different pest and disease categories. It stably identified maize stem borer, slime molds, grey spot, leaf spot, and rust diseases with over 60% accuracy, and red spider and aphid with over 40% accuracy. Field tests evaluated droplet deposition, drift, and pesticide saving rates. The system achieved a minimum droplet coverage of 52% and deposition density of 71.3 drops/cm2, satisfying pest control needs. Pesticide saving and ground wastage rates reached lows of 32.1% and 22%, respectively, significantly reducing overall pesticide consumption and waste. This maize pest and disease detection and precision variable spraying system significantly enhanced identification accuracy, improved pesticide utilization, and reduced environmental pollution, offering a scientific and efficient approach to pest and disease management.