【摘要】 桃树炭疽病和褐斑病具有相似度高、症状关联度高和病斑位置不同等特点,卷积神经网络在识别过程中,卷积层和池化层分别对病害区域进行局部卷积和池化操作,未考虑各病害区域间的上下文相关信息和位置信息,降低了识别准确率。而双向长短期记忆网络由两个正向和反向的长短期记忆网络组成,且各循环单元之间具有反馈连接,能够挖掘和记忆输入序列数据中的上下文相关信息和位置信息。因此,本文提出了一种基于VGGNet-BiLSTM的桃树叶部病害图像识别算法。结果表明,本文提出的算法在测试集上识别准确率为93.73%,具有较高的识别准确率。更多还原

【Abstract】 Peach anthracnose and brown spot have the characteristics of high similarity, high correlation of symptoms and different positions of disease spots. In the process of recognition, convolution layer and pooling layer respectively carry out local convolution and pooling operations on the disease area, and do not consider the context related information and location information among the disease areas, which reduces the recognition accuracy. The bidirectional long-short-term memory network is composed of two forward and backward LSTMs and each recurrent unit has feedback connection, which can mine and memorize the context related information and location information in the input sequence data. Therefore, the paper proposes an algorithm of image recognition of peach leaf diseases based on VGGNet-BiLSTM. The results showed that the recognition accuracy of the proposed algorithm was 93.73% on the test set, which had a high recognition accuracy.更多还原