【作者】 陈月华；

【导师】 胡晓光；

【作者基本信息】 东北农业大学 ， 农业电气化与自动化， 2007， 博士

【摘要】 小麦是世界上最主要的三大粮食作物之一。虫害是影响小麦亩产量的重要原因之一。2005年，我国小麦蚜虫发生面积约为2亿亩，红蜘蛛发生面积约为1亿亩，吸浆虫发生面积约为3000万亩，严重影响了小麦的平均亩产量。有效的害虫监控是提高小麦产量的重要途径之一。小麦害虫的监控、预防工作能够减少、降低虫害的发生及破坏的几率，减少虫害带来的损失，降低小麦生产过程中的成本消耗，提高生产效率。精细农业中的精准施药问题是许多学者正在研究的热点课题，国内、外在这方面的研究尚处于探索阶段。目前为止，在前期的害虫自动识别阶段，以往的研究都是在特定的光源和设定的单一背景下针对静态害虫进行研究的，与实际上的实时识别还有很大的差距。本课题是在国家高科技发展计划(863计划)的基础上提出的，对基于机器视觉的害虫自动检测技术进行了深入的研究：1．设定本文算法的应用对象为非特定环境下的小麦蚜虫若虫，其特点是体积小、颜色与背景(叶片)对比度不强烈，大多有保护色，与背景颜色接近；图像背景为叶片，与实际的处理环境一致；收集了大量的自然光条件下小麦蚜虫的图像，为后续的试验和研究打下了基础。2．阐述了机器视觉及模式识别、图像分类、图像分割的理论基础，分析了多种图像特征，最后确定以纹理特征作为图像的分类基础；在深入研究理论基础上开发出适合本研究的算法，并重点介绍了支持向量机、区域生长、k-均值聚类等分类、分割算法的理论基础，为算法的开发奠定了基础；3．分别对图像的采集、分类、分割以及识别等环节的处理算法进行了研究，开发了在非特定环境(未设定特殊光源、未设定特定的背景)下，适用于害虫自动检测的算法。实现了动态图像的采集和处理，快速的预处理、图像分割和后处理，为进一步的害虫大田实时识别研究做了准备；4．针对样本蚜虫的特点，研究了基于区域生长的害虫图像分割方法、基于支持向量机的分割方法，创造性的将区域生长与支持向量机的方法结合进行害虫图像的分割，通过试验比较，确定了最后的分割方法，针对样本运用算法进行了试验，对试验的结果进行了定性分析，得到了理想的分割结果；5．比较了所提出的分类及分割算法的优缺点，确定了害虫自自动检测模型中所采用的分类、分割的方法，算法不仅应能有效的分割图像，且在速度上能够满足客观环境实时性的要求；6．研究了害虫图像的标记及个数自动统计方法。在识别过程中对害虫数量的统计结果，可以为精准施药提供数据支持，是下一步机械操作的重要参考指标；7．建立了基于视觉的检测系统的模型，在试验室的模拟环境下，完成了图像的采集、预处理、分类、分割以及识别试验，试验结果验证了模型的有效性和可靠性。综上所述，通过理论分析、算法模型的测试等理论和试验相结合的方法，本研究完成了害虫自动检测算法及模型的开发，研究结果对进一步研究药剂的变量喷洒装置、实现精细作业具有重要的理论和实际意义。更多还原

【Abstract】 Wheat is one of the important crops of three in the world. Insect pests of wheat can reducethe output per acre. Aphids happen in our country were 2 hundreds million acres in 2005, and thered spider was 1 hundred million acres. The Attracts the thick liquid insect was 30000 thousandsacres. Those influence the output of wheat seriously.Effective insect pests monitoring is an important way to enhance the output of wheat.Monitoring and prevention can reduce the rate of insect pest’s occur and destruction. Thusproduction cost would be reduced and efficiency would be enhanced during the production.Precise Spraying is a hot topic which many researchers studying it in precision agricultureand in a start stage domestic and overseas. Today, at the stage of insect pests’ automatic detection,all studies were in specific and sole back ground and based on static stage images. Those methodswere far away of real-time detection. This topic was proposed based on 863 plan of China. In thistopic, the author studied thoroughly on automatic detection technology based on machine vision.1. Hypothesis that the object is young wheat insect in non-specific conditions, its haveCharacteristic include that small, color Contrast gradient, and protecting color which is near tobackground; its background are leaves and it is not consistent to processing conditions; A lot ofaphids’ images were collected in natural light and prepared for goes a step of further experimentsand research.2. It elaborates the theory foundation of machine vision, image identification and imagesegment, analysis all kinds of images’ features and making decision of texture as foundation ofimage classification; at the ground of thoroughly theory research, developing the algorithmsuitable this paper. And introducing theory foundations of Support Vector Machine, RegionGrowth, and k-means algorithm and so on to classify and segment images of insect pests, buildingthe foundation for research of algorithms.3. Algorithms of Image capturing, classification, segmentation and pattern processing werestudied respectively. And algorithm was developed suitable for insect pests’ automatic detection innon-specific (undefined light and conditions). Paper finished dynamic images captured andprocessed, faster pro-processing, image segment and last processing, which prepared for goes astep of further in real -time insect pests’ recognition in field.4. Aphids sample characteristics against, it researched the region growth, the Support VectorMachine segmentation and regional growth combined with vector machine segmentation approach,through comparing and testing, final separation methods were selected, and get an ideal segmentation results;5. Comparison the advantages and disadvantages of the proposed classification andsegmentation algorithm, and determine the method of classification and segmentation which cansegment images effectively and its speed can meet the requirements;6. The methods of Image marker and automatic counting the number were proposed. Thenumber of pests can be used for providing accurate data to support machinery application andgiving data support for decision spraying.7. A visual inspection system model was established, at last, this model was tested inlaboratory, and testing results confirmed validity and Reliability of this model;In summary, paper through theory analysis, algorithm model and theory combinedexperiments finished the algorithm and model’s development of insect pests’ automaticdetection,and this result can be used for goes a step of further research variable spraying and hasimportant instrumental and actual meanings for decision work.更多还原