基于改进DeepLabV3+网络的黄瓜叶片病斑分割算法OA

Leaf spot disease is a major factor affecting the quality and yield of the crops such as cucumbers.Accurate segmentation of leaf spots is crucial for precise identification of the diseases and providing farmers with scientific strategies for disease control.Due to the blurred edges and reflective characteristics of leaf spots,existing methods often fail to achieve ideal segmentation results.To address this issue,a cucumber leaf spot segmentation algorithm based on an improved DeepLabV3+network is proposed.First,the original Xception backbone network is replaced with the more lightweight MobileNetV2 network.Secondly,the DenseNet concept is applied to the Atrous Spatial Pyramid Pooling(ASPP)structure,resulting in a DenseASPP(dilated spatial pyramid pooling based on dense connections),which enhances the segmentation performance for multi-scale targets by increasing the receptive field of the network.Additionally,a SENet channel attention mechanism is introduced after DenseASPP to improve the model's feature representation capabilities.Finally,the feature maps extracted by different stages of the backbone network are concatenated with the deep-level feature maps to fully utilize the information contained in the feature maps at each stage.The model was tested and trained on a cucumber leaf disease dataset.The results show that the algorithm achieves 90.55％sensitivity,98.03％specificity,85.43％dice coefficient,and 97.31％accuracy.The segmentation accuracy is significantly higher than that of other mainstream methods,with good generalization ability.This algorithm is applicable to leaf spot segmentation for different crops and can provide valuable reference for crop disease prevention and control.