Online identification and classification of Gannan navel oranges with Cu contamination by LIBS with IGA-optimized SVM

Abstract

Elements such as minerals and heavy metals play important roles in the nutrition and safety of agricultural products. It is necessary to develop rapid, online, real-time and in situ methods for monitoring the substances in farm products. Gannan navel oranges are a unique variety of fruit, which may be affected by Cu pollution due to abundant copper mines and other factors. An online identification and classification system based on laser-induced breakdown spectroscopy (LIBS) was developed to detect possible Cu residue in Gannan navel oranges. First, transmission and classification equipment for Gannan navel oranges was built. Second, an LIBS detection module was designed. Finally, a software system for the whole online detection platform was developed based on the C# programming language. The series of operations for the online detection system, which includes the loading, transmission, detection and classification of orange samples, can be controlled. Since the navel orange has an elliptical shape, the LIBS detection module was designed with a long focal length to reduce the influence of fruit plane size fluctuation. The long focal length was optimized to 698 mm, and the depth of field was ±6 mm. Furthermore, a parameter optimization model using a support vector machine (SVM) based on an improved genetic algorithm (IGA) is proposed to improve the classification effect of Gannan navel oranges. This model avoids the over-learning or under-learning caused by improper parameter selection in the regression prediction of SVM. The IGA is used to optimize the penalty parameter c and the kernel parameter g of SVM. LIBS spectral data from two types of navel orange samples with and without Cu contamination were selected as test datasets, and the classification results were compared with those of the standard genetic algorithm-support vector machine (GA-SVM). The investigation showed that the IGA-SVM can provide better classification of navel oranges based on analysis of the LIBS spectral data, and the classification accuracy can reach 98%, which provides significant guidance for the use of LIBS to quickly realize online screening of heavy metals in agriculture products.