Abstract:Crop disease is one of the most important influencing factors for agricultural high yield and high quality. Accurate classification of diseases is a key and basic step for early disease monitoring, diagnostics and prevention. The optimal individual classifier design is currently the common limitation in most crop disease recognition methods based images. To improve the accuracy and stability of disease identification, a disease recognition method of cucumber leaf images via dynamic ensemble learning was proposed. The approach consisted of three major stages. Firstly, totally 75-dimension color features of leaf image were extracted with image block processing. Secondly, a disagreement approach was used to measure the diversity among 10 classifiers of neural networks with an ensemble technique, where the classifiers were ordered according to the diversity. Finally, with the confidence of classifiers, a classifier subset was dynamically selected and integrated to identify the images of crop leaf diseases. To verify the effectiveness of the proposed method, classification experiments were performed on images of four kinds of cucumber leaf tissues, including 512 samples composed of powdery milder, downy mildew, gray mold and normal leaf. The experimental results showed that the recognition error rate of the proposed method was 3.32%, compared with those of BP neural network, SVM, Bagging and AdaBoost methods, it was reduced by 1.37 percentage point, 1.56 percentage point, 1.76 percentage point and 0.78 percentage point, respectively. The proposed method identified the diseases accurately from cucumber leaf images. Moreover, the method was feasible and effective, and it can also be utilized and modified for the classification of other crop diseases.
