随着大数据、人工智能的发展，农业也逐步趋向智能化，形成了智慧农业。智慧农业主要通过感知、跟踪、监测、预测和数据分析等技术对传统农业进行改造，从而实现农业的智能化决策、精准化生产和可视化管理。在现代农业中，农作物的病害类型主要依靠种植人员的经验来判断，容易出现误判病害类别的现象。这不仅阻碍了农作物种植技术的进步和发展，而且带来了一系列的环境污染问题，因此自动化识别植物病害在智慧农业中至关重要。本文将针对智慧农业下的智慧农业系统中的病害检测需求展开研究。

针对农作物病害类型依靠经验判断不精准的问题，本文以番茄病害为典型，主要研究从番茄病害叶片图像中定位病害类别，利用Mask R-CNN(Mask Regin Connected Convolutional Networks)实现番茄病害叶片的自动化检测和病害区域分割。首先研究了第三章基于Mask R-CNN的番茄叶片病害识别方法，并重点分析了Mask R-CNN算法的核心优势点。然后针对番茄病害叶片数据属性的特殊性，对经典的Mask R-CNN的网络模型进行了两点改进：(1)在Mask R-CNN中增加病害特征筛选模块，减轻了Mask分支负担，在保证检测准确率的基础上，缩短了训练和测试模型的时间，提升了网络的整体效率；(2)改进Mask R-CNN的特征提取网络，选取了ResNet(Residual Network)的优化网络DenseNet(Densely Networks)提取番茄病害叶片图像的特征，相比ResNet它不仅减少了参数量，而且提高了特征图的利用率，从而提高了识别准确率。最后在智慧农业系统的基础上，开发了基于Mask R-CNN的番茄病害叶片识别功能，包括视频图像采集模块，病害识别与分析模块和云平台处理模块。视频图像采集模块包括视频播放器和视频图像截取，实现了从视频中获取目标对象；病害识别与分析模块中包括原图输入，预处理和识别的结果展示，实现了检测番茄病害过程的可视化；云平台处理模块利用FTP协议实现了云服务器与本地计算机之间的文件传输，并利用计划任务自动更新网络模型。为了确保番茄病害叶片识别功能的可用性和稳定性，对以上三个模块进行了功能和性能测试，结果均符合预期。

本文的研究内容具有较高的实用价值，可用于农业种植中的番茄病害叶片识别领域，在一定程度上既可以提高番茄的生产效率和产量，也保护了自然环境。

 

关 键 词：智慧农业， Mask R-CNN， 番茄叶片病害， 病害检测， 卷积神经网络

With the development of big data and artificial intelligence, agriculture is gradually becoming intelligent, and smart agriculture has been formed. Smart agriculture mainly transforms traditional agriculture through technologies such as sensing, tracking, monitoring, forecasting and data analysis. It can realize intelligent agricultural decision-making, precise production and visual management. In modern agriculture, the types of diseases of crops are mainly determined by the experience of planters. It will lead to misjudge the types of diseases. Not only the progress and development of crop planting technology was hindered, but also the phenomenon of drug abuse and misuse was triggered, which a series of pollution problems has been brought. Therefore, automatic identification of plant diseases is essential in smart agriculture. This article will focus on the needs of disease detection in smart agriculture systems under smart agriculture.

 

In view of the problem of inaccurate judgment of crop disease types based on experience. This article takes tomato disease as a typical example and mainly studies the classification of disease types from tomato disease leaf images. The Mask R-CNN (Mask Regin Connected Convolutional Networks) is used to realize automatic detection and disease region segmentation of tomato disease leaves. Firstly, the core principles of several target detection algorithms of R-CNN (Regin Connected Convolutional Networks), Fast R-CNN (Fast Connected Convolutional Networks), Faster R-CNN (Faster Connected Convolutional Networks) and Mask R-CNN are studied. The core advantages of Mask R-CNN algorithm selected in this paper are emphasized. Then, in view of the particularity attributes of tomato disease leaves, this paper makes two improvements to the classic Mask R-CNN network model:(1) Add disease feature screening module in Mask R-CNN. It reduces the burden on the Mask branch and the training and testing time. It improves the overall efficiency of the network on the basis of ensuring detection accuracy. (2) Feature extraction of tomato disease leaves Select ResNet's (Residual Network) optimization network DenseNet(Densely Networks)  to improve the feature extraction network of Mask R-CNN. Compared with ResNet, not only the parameter amount was reduced, but the utilization rate of the feature map was improved. Thereby improving the accuracy of recognition and achieving Expected effect. Finally, tomato disease leaf recognition function based on Mask R-CNN was developed on the intelligent agricultural decision-making system. Including video image acquisition module, disease identification and analysis module, cloud platform processing module. The video image acquisition module can capture the target object from the video, which include a video player and a screenshot function. The disease identification and analysis module realize the recognition function of tomato disease leave, which include the input of original images, the display of preprocessing and identification results. The cloud platform processing module uses the FTP protocol to achieve file transfer between the cloud server and the local computer. And the automatic updating of the model was realized by using the scheduled task. In order to ensure the availability and stability of the tomato disease leaf identification function, the above three modules were tested. The results were in line with expectations.

 

The research content of this paper has high practical value and can be used in the field of tomato disease leaf identification in agricultural planting, which has improved the tomato production efficiency and yield to a certain extent.

 

Keywords: Smart Agriculture, Mask R-CNN, Tomato Leaf Disease, Disease Detection, Convolutional Neural Network

[1] 张丽丽. 关于智慧农业的概述[J]. 新农业, 2019 (20): 20.
[2] S. Savary, A. Ficke, J.-N. Aubertot, and C. Hollier, “Crop losses due to diseases and their implications for global food production losses and food security”[J] 2012 4(519): 1876-4517.
[3] B. Ney, M.-O. Bancal, P. Bancal, I. Bingham, J. Foulkes, D. Gouache,N. Paveley, and J. Smith, “Crop architecture and crop tolerance to fungal diseases and insect herbivory. Mechanisms to limit crop losses” 2013 135(561): 0929-1873.
[4] 关莹.基于图像处理的水稻叶部病害快速识别方法研究[D].东北农业大学,2018.
[5] 任端阳. 我国农业知识产权与智慧农业发展对策研究[D]. 中国科学技术大学, 2017.
[6] 程帅明, 姜鹏飞, 张振, 等. 基于大数据的智能农业传感器设计[J]. 智能城市, 2017, 3(2): 44-46.
[7] 申格,吴文斌,史云,杨鹏,周清波.我国智慧农业研究和应用最新进展分析[J].中国农业信息,2018,30(02):1-14.
[8] 袁小平, 徐江, 侯攀峰. 基于物联网的智慧农业监控系统[J]. 江苏农业科学, 2015, 3: 376-378.
[9] 蔡伟杰. 面向农业植保智能化作业的无人机地面站系统研究[D]. 深圳大学, 2017.
[10] 张松茂. 现代农场智慧农业系统建设研究[D]. 南京邮电大学, 2014.
[11] 张晓朋. 基于 485 总线和虚拟仪器的智能农业监控系统设计[J]. 计算机测量与控制, 2017 (2017 年 02): 85-87,115.
[12] 周新淳, 张瞳, 吕宏强. 基于物联网的精准化智慧农业大棚系统设计[J]. 国外电子测量技术, 2016 (12): 44-49.
[13] 赵胜利. 作物生长感知与智慧管理物联网平台架构与实现[D]. 南京农业大学, 2015.
[14] 岳梦婕. 面向智慧农业种植领域数据处理算法的研究[D]. 浙江理工大学, 2016.
[15] 杨凌雯. 基于数据挖掘的智慧农业生产系统的研究[D]. 杭州: 浙江理工大学, 2016.
[16] 徐润森, 薛艳肖. 基于云计算的智能农业监控平台架构研究[J]. 互联网天地, 2015, 3: 23-28.
[17] 张丽良, 高宏伟, 向小林. 嵌入式系统中农业大棚环境控制技术的研究[J]. 沈阳理工大学学报, 2016, 35(3): 12-16.
[18] Wang Y, Yang X, Chen Y, et al. A biorobotic adhesive disc for underwater hitchhiking inspired by the remora suckerfish[J]. Science Robotics, 2017, 2(10): eaan8072.
[19] 章璐杰. 基于物联网的智慧葡萄园管理系统的优化研究[D]. 浙江大学, 2017.
[20] 林峰峻. 象山县智慧农业综合服务平台设计与实现[D]. 大连理工大学, 2015.
[21] 吴晓行. 南京市智慧农业发展现状及对策探析[J]. 江苏科技信息, 2017 (3): 6-10.
[22] 汪京京,张武,刘连忠等.农作物病虫害图像识别技术的研究综述[J].计算机工程与科学,2014,36(07):1363-1370.
[23] 刘娜,赵慧,包思圆等.基于图像处理的黄瓜叶子病害识别系统的研究[J].山东工业技术,2018,(5):138.
[24] 阎园园. 基于图像处理的番茄病害识别[D].新疆大学,2019.
[25] 郭小清,范涛杰,舒欣.基于改进Multi-Scale AlexNet的番茄叶部病害图像识别[J].农业工程学报,2019,35(13):162-169.
[26] Realization Conference of Robot Revolution. Japan’s robot strategy [R]. Japan: Ministry of Economy, Trade and Industry, 2015.
[27] 赵春江, 杨信廷, 李斌, 等. 中国农业信息技术发展回顾及展望[J]. 中国农业文摘-农业工程, 2018 (2018 年 04): 3-7.
[28] Mao Y, Wang K, Tang C G, et al. Application of internet of things technology in modern agriculture at home and abroad [J]. Jiangsu Agricultural Sciences, 2016, 44(4): 412-414.
[29] 赵春江, 李瑾, 冯献, 等. " 互联网+" 现代农业国内外应用现状与发展趋势[J]. 中国工程科学, 2018 (2018年02): 50-56.
[30] 兰晓东.基于 RGB-D 视觉信息的物体识别算法研究[D].中北大学,2018.
[31] Ahmad I S, Reid J F, Paulsen M R, et al. Color classifier for symptomatic soybean seeds using image processing[J]. Plant disease, 1999, 83(4): 320-327.
[32] Singh V, Misra A K. Detection of plant leaf diseases using image segmentation and soft computing techniques[J]. Information Processing in Agriculture, 2017, 4(1): 41-49.
[33] Zhang S, Zhang S, Zhang C, et al. Cucumber leaf disease identification with global pooling dilated convolutional neural network[J]. Computers and Electronics in Agriculture, 2019: 422-430.
[34] Sladojevic S, Arsenovic M, Anderla A, et al. Deep Neural Networks Based Recognition of Plant Diseases by Leaf Image Classification[J]. Computational Intelligence and Neuroscience, 2016: 3289801-3289801.
[35] Ali H, Lali M I, Nawaz M Z, et al. Symptom based automated detection of citrus diseases using color histogram and textural descriptors[J]. Computers and Electronics in agriculture, 2017, 138: 92-104.
[36] Patil J K, Kumar R. Analysis of content based image retrieval for plant leaf diseases using color, shape and texture features[J]. Engineering in agriculture, environment and food, 2017, 10(2): 69-78.
[37] Zhang F, Li X, Yang H. Image feature extraction and recognition of plant leaf disease in complex background[J]. Journal of Liaoning University (Natural Sciences Edition), 2016 (4): 5.
[38] Mohanty S P, Hughes D P, Salathé M. Using deep learning for image-based plant disease detection[J]. Frontiers in plant science, 2016, 7: 1419.
[39] Mads D,Henrik K,Henrik S M. Plant species classification using deep convolutional neural network[J]. Biosystems Engineering,2016,151.
[40] Simonyan K, Zisserman A. Very deep convolutional networks for large-scale image recognition[J]. arXiv preprint arXiv:1409.1556, 2014.
[41] Szegedy C, Vanhoucke V, Ioffe S, et al. Rethinking the inception architecture for computer vision[C]//Proceedings of the IEEE conference on computer vision and pattern recognition. 2016: 2818-2826.
[42] N. Schor, A. Bechar, T. Ignat, A. Dombrovsky, Y. Elad and S. Berman, "Robotic Disease Detection in Greenhouses: Combined Detection of Powdery Mildew and Tomato Spotted Wilt Virus,"in IEEE Robotics and Automation Letters, vol. 1, no. 1, pp. 354-360, Jan. 2016.
[43] He K,Gkioxari G,Dollar P,et al.Mask R-CNN[J].2017,14(31):424-430.
[44] 郭萧. 基于机器学习算法的智慧农业系统研究[D].西安电子科技大学,2018.
[45] He K,Zhang X, Ren S,et al.Spatial Pyramid Pooling in Deep Convolutional neutral networks for Visual Recognition.[J].IEEE Transactions on Pattern Analysis & Machine Intelligence ,2015,37(9):1904-1910.
[46] Ross G ,“Fast R-CNN”.[C]The IEEE International Conference on Computer Vision (ICCV), 2015, pp. 1440-1448
[47] S Ren, K He, R Girshick, J Sun ,“Faster r-cnn: Towards real-time object detection with region proposal networks”Advances in neural information processing systems, 2015: 91-99
[48] 何维龙. 基于 Mask R-CNN 的无人机影像建筑物检测方法研究[D]. 东华理工大学, 2019.
[49] He K, Zhang X, Ren S, et al. “Deep residual learning for image recognition”[C] Proceedings of the IEEE conference on computer vision and pattern recognition. 2016: 770-778.
[50] Lin T Y, Dollár P, Girshick R, et al. Feature pyramid networks for object detection[C] Proceedings of the IEEE conference on computer vision and pattern recognition. 2017: 2117-2125.
[51] He K,Gkioxari G,Dollar P,et al.Mask-RCNN[J].2017,14(31):424-430.
[52] 白宝林.基于改进的 Mask R_CNN 的车辆识别及检测[D].合肥:安徽大学,2018.
[53] Kim T, Paik J. Adaptive contrast enhancement using gain-controllable clipped histogram equalization[J]. IEEE Transactions on Consumer Electronics, 2008, 54(4): 1803-1810.
[54] Dai J,He K,Sun J.Instance-Aware Semantic Segmentation Via Multi-task Network Cascades Computer Vision and Pattern Recognition.IEEE,2016, 31(11):3150-3158.
[55] Yu Z, Bajaj C. A Fast and adaptive method for image contrast enhancement[C] 2004 International Conference on Image Processing, 2004. ICIP'04. IEEE, 2004, 2: 1001-1004.
[56] Denoeux T. A k-nearest neighbor classification rule based on Dempster-Shafer theory[M]//Classic works of the Dempster-Shafer theory of belief functions. Springer, Berlin, Heidelberg, 2008: 737-760.
[57] Ou G, Murphey Y L. Multi-class pattern classification using neural networks[J]. Pattern Recognition, 2007, 40(1): 4-18.
[58] Xie S,Girshick R,Dollar P,etal.Aggregated Residual Transformations for Deep Neutral Networks [J].2016, 24(31):424-435.
[59] Iandola F, Moskewicz M, Karayev S, et al. Densenet: Implementing efficient convnet descriptor pyramids[J]. arXiv preprint arXiv:1404.1869, 2014.
[60] Huang G, Liu Z, Van Der Maaten L, et al. Densely connected convolutional networks[C] Proceedings of the IEEE conference on computer vision and pattern recognition. 2017: 4700-4708.
[61] Huang G, Liu S, Van der Maaten L, et al. Condensenet: An efficient densenet using learned group convolutions[C]//Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition. 2018: 2752-2761.
[62] Yaguchi T, Nixon M S. Transfer learning based approach for semantic person retrieval[C]//2018 15th IEEE International Conference on Advanced Video and Signal Based Surveillance (AVSS). IEEE, 2018: 1-6.
[63] Everingham M, Van Gool L, Williams C K I, et al. The PASCAL visual object classes challenge 2007 (VOC2007) results[J]. 2007.
[64] Jiang B, Luo R, Mao J, et al. Acquisition of localization confidence for accurate object detection[C] Proceedings of the European Conference on Computer Vision (ECCV). 2018: 784-799.
[65] Lin T Y, Maire M, Belongie S, et al. Microsoft coco: Common objects in context[C] European conference on computer vision. Springer, Cham, 2014: 740-755.
[66] Monteiro F R, Garcia MAP, Cordeiro L C. Boundedmodel checking of C++ programs based on the Qt oss-platformframework[J]. Software Testing Verification & amp; Reliability, 2017.
[67] 微信辅助教学的教学评价及应用[D]. 刘宝鹏.西北师范大学 2018.
[68] Van der Walt S, Schönberger J L, Nunez-Iglesias J, et al. scikit-image: image processing in Python[J]. PeerJ, 2014, 2: e453.
[69] Postel J, Reynolds J. File transfer protocol[J].