基于混合蛙跳算法优化PCNN的马铃薯病害图像分割
Image segmentation method for potato diseases based on pulse coupled neural network with shuffle frog leap algorithm

为优化马铃薯病斑图像特征提取与病害识别的关键步骤——图像分割的精度，保证分割后的图像能够较好地保留原病斑图像的轮廓与细节，采用混合蛙跳算法优化脉冲耦合神经网络（pulse coupled neural network，PCNN）参数，建立一种高精度的用于马铃薯病斑图像分割的混合蛙跳算法（shuffled frog leaping algorithm，SFLA）-PCNN模型，该模型选用图像分割香农熵与图像分割紧凑度的加权和作为适用度函数，对马铃薯晚疫病害图像进行试探分割，分割正确率为95.41%，实现PCNN参数的自适应优化配置，并获得PCNN参数配置方案为：神经元交互连接系数β=0.38、脉冲激励衰减系数aθ=0.24、激励脉冲幅度衰减系数Vθ=0.82。利用优化后的PCNN对马铃薯软腐病、环腐病、银腐病、粉痂病、灰霉病5种病害图像进行分割，分割正确率分别为94.41%、95.69%、93.89%、93.91%和93.21%，平均正确率为94.42%，证明SFLA-PCNN模型能有效地从背景区域提取马铃薯病斑，可用于马铃薯病斑检测。
To reserve the disease image's profile and details, an image segmentation model, shuffled frog leaping algorithm-pulse coupled neural network (SFLA-PCNN), was proposed through Shuffled Frog Leaping Algorithm by optimizing the PCNN neural network parameters for recognition of potato diseases. The weighted sum of information entropy and compactness degree of image segmentation were chosen as fitness function to make PCNN parameters adapt to configuration. The potato late blight image was adopted as trial image to get the best configuration parameters and the segmentation accuracy was 95.41%. The optimal value of β, aθ, Vθ were 0.38, 0.24 and 0.82, respectively. The SFLA-PCNN model was used to segment the five kinds of potato diseases: soft rot, ring rot, silver rot, powdery scab, and gray mold. The accuracy could reach 94.41%, 95.69%, 93.89%, 93.91%, and 93.21%, respectively, and the average accuracy was 94.42%, indicating that the model could be used to extract the lesion from the background effectively for lesion detection of potato.