Study On The Automatic Diagnosis Technology Of The Cucumber Leaf Diseases Based On Computer Vision

Diseases have been the major limiting factors in cucumber production, and a wide range of foliar symptoms occurs which can make diagnosis difficult in field. The main problem associated to the cucumber disease management is that the controlling measures can't be applied at the perfect time due to the absence of the effective early diagnostic technique. By using traditional diagnosis methods, too much time has been spent. Therefore, it is necessary to explore the novel diagnostic methods which can recognize symptoms of the diseases, and quickly to takes steps to control them. The switch to computer disease diagnosis and prediction system has played an important role in agriculture, although this technique is completely dependant on human process. Based on computer vision system, this study investigated technologies of feature extraction and recognition on cucumber leaf diseases digital images by using digital image processing, pattern recognition and artificial intelligence. Progress in this study was summarized as follows:1. Image acquisition system and digital image database were built. The image acquisition system for house operating mainly included the sample stage, light, lighting structure, CCD camera, and system control computer for data acquisition, processing and other necessary components. The inoculation methods, inoculum concentration and inoculation time for 10 diseases, including cucumber powdery mildew(Erysiphe cichoracearum), cucumber downy mildew(Pseudoperonospora cubensis), cucumber brown speck(Corynespora cassiicola), cucumber anthracnose(Colletortichum orbiculare), cucumber scab(Cladosporium cucumerinum)and others, have been found by comparing the symptom development with different treatments. Moreover, the digital image database mainly included about 300 images for the above diseases, have been established based on the data of the diseased leaf samples at the different symptom development stages collected from different varieties and ecological regions, such as Beijing, Shandong, Liaoning, and other Provinces.2. In order to extract the image features for further recognition research, the foundational features-extracting algorithm and program were established for sample images of diseased cucumber leaves, captured by image acquisition system. Feature extraction was investigated by using gray digital image processing, image enhancing, median filting, morphological processing and segregating for cucumber leaf diseases digital images.3. The diseased spots were described and analyzed with proper parameters coming from image features of cucumber leaf diseases. Color characteristic parameters from different periods of diseased samples were used for classification and identification under RGB and HIS models, and stepwise discriminant analysis was used to select significant parameters. The bavesian classifier method was proposed to classify cucumber anthracnose, cucumber brown spot and healthy region. Results showed that the correct discrimination of cucumber anthracnose, cucumber brown spot and healthy region were 96.67%, 93.33% and 100%, respectively. For the testing,the correct discrimination of cucumber anthracnose, cucumber brown spot and healthy region were 83.33%, 80.00% and 100% , respectively. The difference of contrast, energy and complexity were obvious in five kinds of cucumber leaf diseases. Energy, complexity, form factor and entropy could be diseased severity quantification index. Energy, entropy, moment of inertia could be the criterion for having spot or not.4. The improved PCA and BP classifier were used to identify the diseases, and a diagnosis system for recognition to cucumber disease was developed. The result showed that 369 training times were reduced and correct rate rose 5.67 percent. The correct discrimination of five diseases was 93.67% and 74.67% respectively. Functional requirement for this diagnosis system was described. The structure for graphical user interface and function was introduced, and the system could run well.This research provided important basic theories and techniques for further investigation and development of commercial automatic diagnosis system in plant diseases. It also represents an important value for promoting the application of computer vision, artificial intelligence, image processing, and expert system in plant protection.