Black walnut shell and meat discrimination using hyperspectral fluorescence imaging

Black walnuts have a rich and unique flavor and are very healthy for humans to eat. However, the black walnut shell is especially hard and hazardous to the consumer if it is mixed with the meat in the walnut processing plant. Currently, human intervention is still necessary to manually pick up the walnut shell fragments in order to reach the strict USDA regulations. Therefore, there is a need to develop an effective method to automatically detect the walnut shell from the meat. In this paper, a study on black walnut shell and meat classification using hyperspectral fluorescence imaging is reported. The intact black walnuts after harvested were provided by the USDA AMS. A total of four categories were considered including light meat, dark meat, inner shell and outer shell. Samples were scanned by a hyperspectral fluorescence imaging system at 79 different wavelengths ranging from 425 nm to 775 nm with the 4.5 nm increments. The principal component analysis (PCA) was used to reduce the redundancy of the data. Two statistical pattern recognition methods were investigated. The first approach was Gaussian Mixture Model (GMM) based Bayesian classifier, which modeled the walnut hyperspectral data as a pooled Gaussian distribution, and the discrimination among walnut classes was realized through Bayesian classifier given predetermined Gaussian Mixture Model; The second approach was Gaussian kernel based support vector machine (SVM), which sought an optimal low to high dimensional mapping such that the nonlinear separable input data in the original input data space became separable on the mapped high dimensional space, and hence fulfilled the classification among four walnut categories. In addition, cross-validation method was used to evaluate robustness of proposed classification methods. The experiment results showed the effectiveness of proposed approaches in the application of walnut shell and meat classification, and an overall recognition rate was achieved up to 95.6%.