Research On Potato Disease Recognition Based On Computer Vision Technology

Potato is an important food and vegetable crop,which is another staple food after wheat,rice and corn.It has the characteristics of strong adaptability,high output,and rich nutrition,making it not only a major food on the world food market,but also an important industrial raw material,which has great development value.China is the largest production country of potatoes.Gansu Province has become one of the major potato production areas due to its unique natural resources,farmland climate and environmental resources.Its planting area and output rank first in the country,promoting the development of potato industry in the whole province,which is an important way to increase agricultural income,increase the efficiency of enterprises and make farmers become rich.China is a large agricultural country and a country with many agricultural pests which is not only in variety and widely distributed,but also has complicated disaster conditions.Potato disease has always been one of the major factors that constrain China's sustainable development,while causing economic losses,it also causes environmental pollution and food pollution,and also threatens human food safety.In order to solve the problems of serious potato disease,easy to spread,lack of pest experts and so on,to help grass-roots technicians and the vast planting users to correctly diagnose the potato disease timely and accurately,grasp the occurrence of the dynamic,and guide the prevention and control work,on the basis of computer vision technology,combining intelligent optimization algorithm and deep learning algorithm,a series of studies have been conducted on the automatic identification of potato disease images.The main work is as follows:1.Between July and September of 2015 and 2016 respectively,the collection work of potato disease images was carried out in the main planting area of potato in Gansu Province,Dingxi City “modeling agricultural demonstration area in the national modern agricultural demonstration zone-the demonstration base of Xiangquan Town in Anding District of Dingxi City”,and Lintao county,Kangle county,Min county,Weiyuan county,Zhang county and Longxi county of Dingxi City.It focused on collecting diseased images of potato leaves,then used computer vision technology to conduct preprocessing of cropping images and adjusting the image resolution in the Adobe Photo-shop CS6 x64 Simplified Chinese version software;conduct preprocessing of image graying,gray image median filter and color image median filter in the MathWorks MATLAB R2016 a x64 Simplified Chinese version software.2.Based on the study of the OTSU threshold segmentation algorithm and the SFLA intelligent optimization algorithm,the simple and efficient features of OTSU algorithm are combined with the features of high-efficiency computational performance,the inability to fall into local optima,excellent global search capabilities,and easy-to-implement of SFLA,a OTSU-Shuffled Frog Leaping Algorithm(refers to as as OTSU-SFLA algorithm)based on shuffled frog leaping algorithm was proposed for the first time.Firstly,the preprocessed potato disease image was grayed and the histogram was calculated;then used OTSU algorithm to perform basic segmentation of the disease image based on the histogram;finally,the segmentation result of the OTSU algorithm was used as the starting point for the optimization of SFLA algorithm;based on the segmentation of OTSU algorithm,the SFLA algorithm was used to further optimize the segmentation and optimization ability with its powerful optimization and calculation capabilities.Used this algorithm to program in the MathWorks MATLAB R2016 a x64 Simplified Chinese version software,and the disease speckle area was successfully obtained from the complex potato disease images and the accurate segmentation of image disease speckle was achieved.Meanwhile,in this algorithm,the concept of image segmentation compactness and calculation method were proposed for the first time,and the weighted sum of image segmentation compactness and cross-entropy was used as the fitness function of the algorithm,after introducing the fitness function in OTSU-SFLA algorithm,not only the algorithm converges quickly,but also it is not easy to fall into the local optimum.3.Based on the research of artificial neural networks and deep learning algorithms,the Convolutional Neural Network was fully used in the field of computer vision outstanding performance as well as its strong modeling capabilities,characteristics of the learning capabilities and pattern recognition capabilities.Based on Caffe open framework,a 13-layer potato disease recognition model based on deep Convolutional Neural Network was constructed.,which consist of one input layer,five convolution layers,three pooled layers,two fully connected layers,one Softmax layer,and one output layer.The ReLU function,as its activation function,integrated feature extraction directly into the learning and training of the model,and combined the feature extraction and classification recognition of the disease images.Restored the binary image after segmentation by OTSU-SFLA algorithm to a color image,then divided it into two groups of samples,each of which rotated counterclockwise every 45 degrees for a total of 7 rotations.After enlarging 8 times of the original samples,the model was trained and adjusted its parameters;another group of samples were used to identify and test the model,and the average accuracy was 95.17%,which showed that the feature extraction and disease identification of potato disease images were well achieved.