Research On The Cucumber Leaf Diseases Recognition Based On Image Processing

Vegetable production is an important channel for many farmers who get rich out of poverty, but the disease is an important factor to affect the yield and quality. The traditional disease recognition methods’ subjective, limitations and fuzzy serious influence the accuracy of the identification, which lead to the wrong diagnosis and delay diagnosis of the disease.With the development of Internet of Things technology, crop networking disease recognition has become a new hotspot and it brings the possibility to crop diseases intelligent, accurate identification. With computer image processing technology and support vector machine, this study take cucumber leaf disease as the example, pretreatment disease image and extract the disease leafs color, texture and shape features, then use computer recognition method to train the classification model and to achieve three kinds of common diseases of cucumber image recognition. The method is scientific, accuracy and has certain guiding significance for Internet of Things in the vegetable diseases networking intelligent recognition.The main contents of this study include the following sections:(1) Preprocessing for disease image. This study use the weighted average method to gray scale processing color images and median filtering method to suppress noise of disease image, and Image enhancement by histogram equalization. Through a series of pre-treatment the image shows a better effect.(2) Segmentation for disease image. First, this study selects trial threshold segmentation for cucumber leaf lesion. Second, this artical fill holes processing and remove unrelated small objects. Finally, the paper use mathematical morphology processing to obtain ideal lesion segmentation image.(3) Feature extraction for disease image. This study extracts each component of the R, G,Iin the HSI color space as the color feature and GLCMcontrast, correlation, energy, homogeneity and entropy as texture features and elongation, complexity, eccentricity, rectangle, area unevenness ratio, roundness as the shape feature. There are a total of14features. (4) Training and testing based on SVM classifier model. Using support vector machine Classification multiple randomly, this study select the training set and the testing set for identification. Unilaterally from the color, texture, shape recognition, the average correct recognition rate is72.23%,90.70%,90.24%; Then three kinds of color, texture and shape features comprehensive recognition, the average correct identification rate is96.00%, which achieve ideal effect of recognition and the recognition rate than a single feature such as color, texture or shape is higher by23.77%,5.93%and5.76%, respectively.From the above results, we can see that the innovation of this paper is as follows:(1) By analysis of each component of the RGB and HSI color space and combination of two color space, this paper extract the R, G, I as the color characteristics, because these components figure is clear. Besides, to achieve a better recognition effect, the comprehensive utilization of color, texture and shape three types of characteristics to identify in this paper.(2) Several randomly selected from the training set and test set for training and recognition, and the average of results is calculated with several experiments, making recognition results more reliable and closer to the true recognition rate than the results only after one training session.(3) In terms of diseases image segmentation, through filling the holes and removing irrelevant small objects, segmented image of the cucumber lesion is more desirable.