Image Information Contribution Evaluation for Plant Diseases Classification via Inter-Class Similarity

Combineingplant diseases identification and deep learning algorithm can achieve cost-effective prevention effect, has been widely used. However, the current field of intelligent plant still faces problems insufficient data inaccurate classification. Aiming to resolve these problems, present research proposes an image information contribution evaluation method based on analysis inter-class similarity. Combining this with active selection strategy provide guidance for collection annotation datasets diseases, so as improve recognition effect reduce cost. The proposed includes two modules: inter-classes similarity module module. images located decision boundary between high classes will be images, they more In order verify effectiveness method, experiments were carried fine-grained classification dataset tomato diseases. Experimental results confirm superiority compared others. This is in disease For detection segmentation, further advisable.