Multi-instance graph approach to wheat leaf disease segmentation

Image segmentation is one of the difficult issues in crop disease recognition. Traditional methods based on threshold or clustering are mostly used in present research, but the precision is low. A multi-instance graph model was proposed in this paper to improve the segmentation performance of crop disease. The image segmentation was formulated as a graph segmentation problem under multi-instance learning framework, where the neighborhood information of pixels was fused to build the feature space and the adaptive geometric relationship between two packages of instances was introduced to wheat leaf disease segmentation, which was combined with the global and local features and had better measurement of interior difference and exterior difference. The experimental results show that this approach outperforms other methods and is effective for wheat leaf disease segmentation.