Study on bloody clam population identification based on multi-spectral image

An identification method was proposed for distinguishing the bloody clams from different populations based on multi-spectral image technology. Three populations of bloody clams were collected from Korea, Shandong, and Zhejiang, respectively. The multi-spectral images of bloody clam shells were acquired by CIR MS3100 multi-spectral camera. The graylevel co-occurrence matrixes (GLCM) of the three sub-images were calculated and the texture features of images were calculated according to the GLCM. 3 ratios were extracted after standard deviation filtering and threshold segmentation for each image. Totally, 15 features were obtained for one sample. 3 principal components were selected by using principal component analysis (PCA). Discriminant analysis and least square-support vector machine (LS-SVM) were used to establish the discrimination models. The prediction accuracy of discriminant analysis and LS-SVM were 64.44% and 46.67%, respectively. The prediction accuracy was low, but it provided a new approach for nondestructive identification of bloody clams from different populations.