Research On Category Identification Of Larvae Based On Near-infrared Spectroscopy And Crochet Image Feature

The plant quarantine in China faces enormous challenges due to thespread of pests in recent years. This paper proposes a novel method for larvaecategory identification by fusing near-infrared spectroscopy feature and crochetimage feature, which combined with near-infrared spectroscopy technology, machinevision, image processing, pattern recognition and knowledge of larvae classification.This method provides more accurate and reliable basis for larvae categoryidentification. This paper mainly contains larvae near-infrared spectroscopy, crochetimage acquisition and preprocessing, spectral matrix modeling and model validation,crochet feature extraction and classification. More details are as follows:(1) Build the real-time acquisition platform of near-infrared spectroscopy toobtain the original spectral data. Then preprocessing will be the next step. Themethods of preprocessing mainly contain: smoothing+first derivative, smoothing+SNV, smoothing+MSC, etc. The preprocessing lays foundation for subsequentmodeling of qualitative analysis.(2) Supervised and unsupervised methods of pattern recognition are using tocalibrate spectral matrix. Supervised method is as follows: PLS-DA and SIMCA,unsupervised method is as follows: discriminant cluster analysis of Mahalanobisdistance and correlation coefficient.(3) Build the real-time acquisition platform of image to obtain crochet image.The methods of crochet image preprocessing mainly contain: image smoothing,image segmentation, etc. The preprocessing lays foundation for subsequent crochetimage processing.(4) Feature extraction algorithms of morphological skeleton extraction, Harris,Hough and curve fitting are adapted respectively in this paper. After comparison andanalysis, the results prove that:①Harris algorithm have the translation androtation invariance.②Hough and curve fitting algorithm can reflect the crochetsort intuitively.(5) Database of spectral feature and image feature are established, in which contain larvae category: Heliocoverpa armigera, Spodoptera exigua, Prodenia lituraand Ostrinia nubilalis. Select the BP neural network structure to achieve larvaeidentification. Use information fusion technique for two kinds of information, i.e.near-infrared spectroscopy and machine vision. At last compare the accuracy ofidentification of different model.