Research On Pest Detection In Fruit And Vegetable Using Hyperspectral Imaging Technology

Fruit and vegetable trade occupies important status in our country agricultural product import and export trade. It’s a big issue that fruits and vegetables are easy to carry harmful insects. For the drawbacks of the conventional testing, the hyperspectral imaging technology,which can realize the rapid and nondestructive detection of harmful insects in fruits and vegetables, is introduced into the field of inspection and quarantine insect pests in vegetables and fruits. This study is useful for improving the level of inspection and quarantine technology of agricultural products in China, which can protect the good development of agricultural production of our country and people’s health and the national economy.To achieve automatic detection of the fruit and vegetable pests, this research focuses on the key point of selecting the region of interest(ROI) through automatic method, feature generation and fusion, wavelength selection, as well as establish the testing classification model. The main content of this paper is as follows:The hyperspectral reflectance imaging technology was applied to the apple multilevel classification. First, the image reflectance spectra of hyperspectral images were extracted as classification features for 3 grades(normal apples, bruise apples and pest infestation apples),partial least squares discriminant analysis(PLSDA) was used to develop the classification models to discriminate the three types of apple simultaneously using reflectance spectra. The hyperspectral image has many wavelengths, and different wavelengths exist correlation and redundancy. Therefore the sensitivity wavelengths were optimized and selected by using principal component analysis(PCA) and reflectance sensitivity analysis(RSA) method. The results showed that the high spectral reflectance image technology can implement the grade classification of apples effectively, and this technique provided a new feasible implementing scheme for food classification recognition detection.The hyperspectral reflectance imaging technology was applied to the internal pest detection of hawthorn. The hyperspectral image characteristic of hawthorn was difficult to accurately extract, because hawthorn texture was relatively coarse. This study took the different approachs of feature extraction and integration to improve the detection accuracy of hawthorn internal pest. The partial least-squares discriminant analysis model evaluated the stability of optical characteristics extraction, chemical information extraction and texture features extraction methods by hyperspectral image. Finally, this research realized the accurate detection of hawthorn internal pest with feature fusion algorithm.The hyperspectral transmission imaging technology was impressed on the detection of pod borer inside vegetable soybeans. Region of Interest(ROI) is the image area which can cause user’s interest and reflect the image content. Considering the characteristics of the transmitted light intensity, the automatic extraction method of ROI was introduced in the study to achieve a complete extraction of beans. After that, this study established the etiella zinckenella detection model by extracting the entropy, energy, and mean of the ROI. The results turned out that hyperspectral transmission imaging technology was useful for detection the etiella zinckenella inside vegetable soybeans, and the operation was simple, goodreal-time, high reliability, it provided a strong support for real-time online inspection of hyperspectral imaging.