Abstract:

Anthracnose is a pervasive and serious disease in citrus orchards. In order to improve the accuracy and efficiency of disease identification under orchard environmental conditions and ensure fruit yield and quality， this study recognized the ROI fusion features of diseases image in orchard. A dataset comprising of 9 types of citrus anthrax images depicting various disease sites and stages was collected for model training purposes. In the disease ROI feature extraction and detection module， image color， texture features， and their fused features were extracted to obtain more disease feature information， and form an SVM classifier. The trained SVM classifier was used to detect and identify the disease images to be tested. The trained SVM classifier successfully detected and recognized the target disease images by fusing spectral and texture features， the average accuracy rate of disease identification can reach 94%， with an average processing time for disease identification of 0.005 s. This method had high accuracy and strong robustness for the detection and recognition of citrus anthracnose in complex natural environments， and was of great significance for the prevention and control of citrus diseases.

Key words: anthracnose, deep learning, object detection, classification recognition, disease diagnosis, SVM

摘要：

炭疽病在柑橘园普遍发生、危害严重，为提高果园环境条件下病害识别的及时性和准确率，保障果品产量和品质，对果园环境条件下病害图像的ROI融合特征进行识别。收集果树不同发病部位、病害不同发病阶段的9种类型的柑橘炭疽病害图像作为模型训练的数据集；在病害ROI特征提取检测模块中对图像颜色、纹理特征及其融合特征进行提取，以获得更多的病害特征信息，并形成SVM分类器；使用训练好的SVM分类器进行待测病害图片的检测识别。将光谱特征与纹理特征融合送入训练好的SVM分类器，病害识别准确率平均可达94%，病害识别平均用时0.005 s。该方法对复杂自然环境下柑橘炭疽病的检测识别具有较高的精准度和较强的鲁棒性，对柑橘疾病的防控具有重要意义。

关键词: 炭疽病, 深度学习, 目标检测, 分类识别, 病害诊断, SVM