Machine learning for detection of walnuts with shriveled kernels by fusing weight and image information

摘要

Walnut is one of the popular nut foods with rich nutritional value and medicinal value. However, it is difficult to detect the internal quality of walnuts because of their solid shell. In this study, a novel method was proposed to nondestructively detect the shriveled kernels in shelled walnuts based on the fusion of image and weight information by machine learning. First, the image and weight information of walnut samples was collected using an industrial charge-coupled device camera and an electronic balance. Then, three kinds of models including partial least squares-linear discrimination analysis, a support vector machine (SVM) and a particle swarm optimization algorithm with back propagation (PSO-BP) were established to discriminate walnuts with shriveled kernels. The classifying effectiveness of all methods was comprehensively compared to determine the optimal one. Finally, the testing results were used to evaluate the three models. Under the same conditions, SVM has the best performance. The classification accuracy and average costing time of SVM were 97% and 0.001 s. Overall research demonstrated that the machine learning method based on weight and image information can be used to quickly, accurately and nondestructively detect the walnuts with shriveled kernels. Practical Applications Nondestructively detection of walnuts has significant value for walnuts processing in practical application. It can allow the walnut industry to provide better-tasting walnut to the consumers, and thus, improve industry competitiveness and profitability. A strategy for detecting walnuts with shriveled kernels was proposed based on the fusion of weight and image information using machine-learning algorithms. The SVM model can quickly and accurately classify walnuts with shriveled kernels using information fusion of imaging and weighing. This work is valuable for online sorting of walnuts with shriveled kernel.