基于CT图像的苹果苦痘病与磕碰伤识别

对苦痘病进行持续、准确、量化的无损检测，以及育种专家对新品种苹果的抗苦痘病表型研究，都需要苦痘病准确识别技术的支持。针对磕碰伤对苦痘病识别产生干扰，降低了识别准确率问题，基于苹果CT图像，提出了一种苹果苦痘病和磕碰伤识别方法。首先，采用最大类间方差法、区域标记、中值滤波等方法，对337帧苹果CT图像进行图像分割和伤病区域定位；其次，对伤病区域进行特征提取，提取其形状特征、纹理特征和位置特征共18种特征信息；然后，利用多元逐步回归和类距离可分离性判据2种方法分别选取特征信息，将2种方法选出的相同特征作为本文的选用特征信息；最后，分别使用遗传算法优化的支持向量机和默认参数的支持向量机，对苹果苦痘病和磕碰伤进行识别。识别结果表明，经过遗传算法优化的支持向量机的总体识别准确率高于93%，默认参数的支持向量机算法的总体识别准确率高于84%。遗传算法优化后的支持向量机的识别准确率明显优于默认参数的支持向量机的识别准确率。

Recognition of Apple Bitter Pit and Bruise Based on CT Image

Continuous, accurate, and quantitative non-destructive testing of bitter pit, as well as research on the phenotype of new varieties of apples by breeding experts, require the support of accurate bitter pit identification technology. The presence of bruise will interfere with the recognition of bitter pit and reduce the recognition accuracy. Therefore, it is necessary to carry out research on the recognition of bitter pit and bruise. Based on the CT images of apples, a method for identifying apple bitter pit and bruise was proposed. The method such as maximum between-class variance, region labeling and median filtering were used to segment 337 apple CT images and locate the injured area. Following this step, a total of 18 features of the shape, texture and location of the injured area were extracted. Additionally, the feature information was selected using two methods of multiple stepwise regression and class distance separability criterion. The common features selected by the two methods were used as the selected feature information. Finally, the support vector machine optimized by genetic algorithm and the support vector machine with default parameters were used to identify apple bitter pit and bruise. The recognition results showed that the overall recognition accuracy of the support vector machine optimized by the genetic algorithm was higher than 93%, and the overall recognition accuracy of the support vector machine algorithm with default parameters was higher than 84%. The recognition accuracy of the support vector machine optimized by the genetic algorithm was obviously better than that of the support vector machine with default parameters. The research results can be used to cultivate the phenotype analysis of apple bitter pit and bruise.