基于区分深度置信网络的病害图像识别模型

对枸杞病害进行及时、准确地检测识别对于病害的监测、预测、预警、防治和农业信息化、智能化建设具有重要意义。研究提出了一种基于区分深度置信网络的枸杞病害图像分类识别模型。首先，把枸杞叶部病害图像通过自动裁剪方式获得包含典型病斑的子图像，再采用复杂背景下的图像分割方法分割病斑区域，提取病斑图像的颜色特征、纹理特征和形状特征共计147个，结合区分深度置信网络和指数损失函数建立了病害识别模型。实验结果表明，该方法对于病害图像识别效果较好，与支持向量机相比，基于区分深度置信网络的病害图像识别模型高效地利用了底层图像特征的高层表示，解决了没有足够标注数据时的图像识别问题。

Recognition model of disease image based on discriminative deep belief networks

To detect and identify the disease of Chinese Wolfberry in time and accurately is very important on the disease monitor, prediction, early warning，treatment and the construction of agricultural information and intelligence. The deep architecture of disease image classification and identification is proposed based on discriminative deep belief networks. First of all, this paper automatically crops the leaf disease image of Chinese Wolfberry into the sub-image containing typical spots, and then researches segmentation under complex background and the image feature extraction, the features is a total of 147 on color feature, texture feature and shape feature. Disease recognition model is established with discriminative deep belief networks and exponential loss function. Experimental results show that, the method has good effect on  image recognition. Compared with the support vector machine, the disease image recognition model based on discriminative deep belief network not only can effectively use the high-level representation of low-level image features but also can solve the problem of data annotation image recognition.