基于支持向量机的苹果检测技术

Detection on defects of apples based on support vector machine

由于苹果果梗和缺陷的识别是苹果检测中的难点,两者的误分类会造成苹果等级的误判.作者提出了苹果果梗和缺陷图像分形特征提取的改进算法,构建了支持向量机并采用SMO算法对其进行训练.用计算机视觉系统采集苹果图像,然后提取苹果果梗和缺陷的分形特征作为支持向量机的输入进行识别.用富士苹果进行试验,得到的平均识别正确率为90·6%.

Identification of stem and blemish is a thorny problem in apple grading. If the stem is incorrectly classified as blemish, a false grade will be assigned to the fruit. A new method based on support vector machine （SVM） is proposed to identify blemish and stem on Fuji apples. A fractal algorithm was adopted and modified to extract features of stem and blemish. The SVM was constructed and trained using sequential minimal optimization （SMO） algorithm. The fractal features of stem and blemish were fed as input of the SVM to distinguish stem and blemish. The test results on Fuji apples showed that an average of 90% classification accuracy was achieved by using the proposed method. In a more general way, the proposed method is applicable to feature detection for other types of produce.