Research of Diagnosing Cucumber Diseases Based on Hyperspectral Imaging
|School||Shenyang Agricultural University|
|Course||Applied Computer Technology|
|Keywords||Hyperspectral imaging technology Cucumber Diseases Principal component analysis Feature extraction Support vector machine|
The cucumber is the most important vegetable in China, which are planted widely and get the economic benefits significantly. With the development of greenhouse vegetables, the cucumber in the "Vegetable Basket Project" plays an important role especially. However, the cucumber are affected by various factors and diseases in the process of growth frequently, thus affecting the yield and quality of cucumber, or even crops. The use of pesticides receive good results sometimes, but leading to the pollution of the cucumbers and the environment. The effective technique to solve such problems is real-time monitoring on cucumber, early warning and spraying pesticides precisely, and the most important premise is to get the information of infected cucumbers rapidly and accurately. With the development of modern information science and technology, the technologies of image processing and recognition, spectral analysis techniques have been applicated in the diagnosis of plant diseases, and offer a powerful means of diagnosis for the realization of plants fast, accurately and nondestructively.Hyperspectral imaging technology as a new agricultural technology for crops disease detecting, discusses the feasibility of cucumber disease detection in this paper. Due to high-dimensional data of the hyperspectral sensing images brings difficulties for further processing, to solve this problem, this paper offers a principal component analysis method for dimensionality reduction to detect cucumber downy mildew. Firstly the collected hyperspectral image data is disposed through principal component analysis. Optimized the three wavelengths 634 nm,679 nm and 700 nm. After the feature images are pretreatmented, the features are extracted by gray statistics,color in the two areas. Three wavelengths will each feature image by pretreatment statistics from the gray scale, color made two feature extraction, twenty-two eigenvalues are extracted initially, and then the nine eigenvalues are optimized and selected by discriminant analysis, the most representative parameters are selected. Finally, support vector machines and neural network technology, the sample from the number of samples of cucumber diseases, feature images, the characteristic parameters and kernel function to classify and compare, identify the best parameters.The paper is carried out by image acquisitions,the principal component analysis,image preprocessing,feature extraction and pattern recognition methods,and made a better results.Paper conclusions will promote the technique of hyperspectral imaging and the application will have some references in other crop diseases of early diagnosis.