基于图像识别的向日葵病害诊断技术研究

China is one of the world biggest countries in term of sunflower production which ranked sixth in the world and has become the world's largest country of the edible sunflower cultivation. Furthermore, Inner Mongolia is a province which has the most prosperous sunflower cultivation and now becomes the main area of Chinese Sunflower’s District. It is available of the whole sunflower to be as medicine or food and also it has other economic worth. However the diseases reduce the production and the quality seriously. Traditional quantitative classifications for such diseases are mainly by manual measurement or visual inspection of the professional which cost a large amount of time and energy and seems inaccuracy either. With the extensive developing of Image Recognition in the field of agriculture, how to diagnose the plant diseases and insect pests turns into a research hotspot. It is feasible for image processing and recognition to diagnose the sunflowers..This Research regard sunflower diseases as research object, combining with sunflowers’ disease characteristics and SURF Feature analysis method, adopting automated visual inspection with work smart technology to illustrate the new method and workflow of which from sunflower plant diseases image acquisition to classification of its plant diseases. So as to achieve the promoting gradually from laboratory research to practical application field steadily. In addition, it provides new thought and method for rapid diagnosis of other crop diseases .

我国是向日葵生产大国，排在世界第六位，并已成为世界上种植食葵面积最大的国家。内蒙古自治区是当前我国向日葵种植面积最大的省区，并成为“中国向日葵带”的主要地区。向日葵浑身是宝，既有药用价值又有食用价值，还有其它方面的经济价值，但由于病害严重影响了向日葵产量和质量。传统对病害的量化分级主要靠专业人员手工测量或目测，费时、费力且不准确。随着图像识别技术在农业领域应用中的广泛开展，如何能快速有效地进行病虫害诊断成为了一个研究热点。通过图像处理、图像识别可进行有无病害的诊断。.本研究项目以向日葵病害为研究对象，结合向日葵病害特点及SURF特征的分析方法，结合计算机视觉技术和人工智能技术，阐明了从向日葵病害图像采集到病害分类识别的新方法、工作流程，并逐步实现从实验室研究到大田实际应用的稳步推进，为其它农作物病害的快速诊断研究提供新的思路和方法。

向日葵病虫害呈逐年上升趋势，严重影响到其产量和质量，给以向日葵作为重要经济和油料作物的北方地区带来重大损失，因此对向日葵病害的诊断识别是目前我区迫切需要处理的一大难题。传统的病害诊断方法主要是植保人员通过肉眼识别，并结合植物病害的病原菌的形态进行判断，这种方法诊断效率低，很难及时、准确的判断病害类型。科学技术的进步使图像处理与识别技术向人工智能逐步过渡，随机森林法、BP神经网络识别植物病害已得到广泛利用。本课题主要采用图像处理技术结合随机森林法、BP神经网络进行向日葵叶部病害识别，通过对神经网络的优化提高病害图像识别准确率。因此，本课题的研究具有现实意义。.本课题主要以向日葵白粉病、细菌性叶斑病、黑斑病、霜霉病、褐斑病为研究对象，采用随机森林法对病害进行诊断识别，利用BP神经网络进行病害诊断，并通过BP神经网络的优化提高识别正确率。用遗传算法优化该网络的初始权值和阈值，稳定网络性能，降低误差，提高识别率；在Matlab环境下结合不同用户使用的情况，利用GUI搭建基于随机森林法的向日葵叶部病害识别系统、GA-BP神经网络识别向日葵叶部病害系统。.深度学习作为新兴热门算法，课题搭建经典目标检测深度学习模型YOLOv3，完成对该模型的训练和识别实验，分析其与传统机器学习算法相比各自所具有的优点和不足。.完成了算法的树莓派移植，在树莓派环境下安装配置OpenCV，TensorFlow，搭建pyqt5用户界面，实现算法在移动终端下的使用，以检测算法的实际可用性。