Abstract:

Aiming at the problems of large workload of pumpkin diseases identification, high difficulty in diseases identification and low pesticide utilization rate, this paper proposes an identification method of pumpkin leaf diseases based on K-means clustering and LBP feature, which provides theoretical basis and technical support for precision application of intelligent robots. This method segments pumpkin leaf diseases spots based on K-means clustering and removes noise through morphological treatment. Then, diseases spots sampling area is demarcated and LBP operator is used to extract the feature of diseases spots. Finally, gray-scale map and LBP feature map of sampling area are input into two-channel convolutional neural network for feature extraction, and weighted fusion classification network is used for feature fusion and Softmax is used for pumpkin leaf diseases identification. The experimental results show that the pumpkin leaf disease identification method proposed in this paper can identify leaf spot, powdery mildew and downy mildew with high accuracy, and its performance is better than the single-channel CNN disease identification method using gray-scale map of leaves, which meets the precision application requirements of spraying pesticide robots and means that this method is beneficial to the control of pumpkin diseases.

Key words: pumpkin diseases, spraying pesticide robots, K-means clustering, Local Binary Pattern（LBP） feature, convolutional neural network

摘要：

针对南瓜病害识别工作量大、病害甄别难度高和农药利用率低等问题，提出一种基于K-means聚类与LBP特征的南瓜叶片病害识别方法，为智能机器人精准施药作业提供理论依据与技术支撑。该方法基于K-means聚类分割南瓜叶片病斑并经过形态学处理去除噪声，然后标定病斑采样区计算病斑LBP特征图，最终经由双通道特征提取网络及特征融合网络完成对病斑全局特征与细节特征的提取并使用Softmax分类器进行南瓜叶片病害识别。实验结果显示，提出的南瓜叶片病害识别方法能够以较高的准确率识别叶斑病、白粉病及霜霉病，性能优于采用病斑灰度图和LBP特征图的单通道CNN病害识别方法，满足施药机器人精准施药作业要求，利于南瓜病害防治工作。

关键词: 南瓜病害, 施药机器人, K-means聚类, 局部二值模式（LBP）特征, 卷积神经网络