论文部分内容阅读
储粮害虫特征选择是粮虫图像识别中一个关键的预处理环节。提出基于v折交叉验证训练模型识别率和所选特征个数的特征子集评价准则,将人工免疫算法应用到粮虫的特征选择。该算法从粮虫的17维形态学特征中自动选择出面积、周长等7个特征的最优特征子空间,采用参数优化之后的SVM分类器对90个粮虫样本进行分类,识别率达到95.5%以上,并与PCA法、GA法和原始特征法进行了对比,结果表明人工免疫算法降低了特征空间的维数,提高了分类器的识别率,证实了基于人工免疫算法的粮虫特征选择是可行的。
The characteristic selection of stored grain pests is one of the key pretreatment links in image recognition of pests and insects. The criterion of feature subset evaluation based on the v-fold cross-validation training model recognition rate and the number of selected features is proposed. The artificial immune algorithm is applied to the feature selection of the worm. The algorithm automatically selects the optimal feature subspace of seven features, such as area and perimeter, from the 17-dimensional morphological features of the grain insects, and classifies 90 grain insects using the SVM classifier after parameter optimization, and the recognition rate reaches 95.5%. Compared with the PCA method, the GA method and the original feature method, the results show that the artificial immune algorithm reduces the dimension of the feature space and improves the recognition rate of the classifier, confirming the characteristics of the grain insects based on the artificial immune algorithm The choice is feasible.