Abstract:

【Objective】 The objective of this study is to establish a kind of quality nondestructive testing method, which can be used for grading peanut quality, based on computer vision. 【Method】 Digital color image of peanuts were taken and scanned from 2 side faces each of 100 kernels each of 11 different kinds of quality and of 100 kernels each of 100 groups. Referring to national standards of China, 11 restrictive items of peanuts kernels of different kinds of quality have been devised and quantized. Also the distinguishable methods of size and grades of peanuts have been devised. Fifty-four appearance characters belonging to 3 categories of shape, color and texture had been measured. And then the characters were optimized based on PCA. ANN and SVM quality testing models were built and compared. Using MATLAB and SPSS, the results were analyzed. 【Result】 The SVM model based on the first 16 PCs could detect at 95% accuracy different qualities of unsound/mildew/impurity/different peanut varieties. Also these results fitted at 93% accuracy close to that of tested by manual. By testing 100 groups of peanuts, the correct rate of size and grade was 92%.【Conclusion】 The result of this study has provided a new method which can be used in peanut quality testing and grade testing, and this method is good and stable. This method can be generalized and used in peanuts testing of different qualities, grade screening, processing, and commodity grading and pricing.

Key words: peanut kernel, computer vision, quality grades, nondestructive detection