Research On Chemometrics Methods For Classification Of Seed By Hyperspectral Imaging

Seed is the most basic and important production material in the progress of agricultural production. Seed also is a carrier for many agricultural technologies and agricultural means of production, which has played a key role to production and income of agriculture. Thus, the quality safety of seed is very important to the development of agricultural production. Seed purity reflects the degree of seed varieties in typical consistent characteristics, so it is great important to improve the reliability and accuracy of seed purity detection for guaranteeing the quality of seeds. Due to a variety of factors, mixed seed is an inevitable phenomenon in the progress of production seed, which pose a serious threat to the interests of the farmer. Therefore, how to improve the veracity and reliability of seed purity testing for ensuring the quality of seed and agricultural product fruitful has important research significance.Chemometrics method, integrated the application of statistics, mathematics and computer science and other related disciplines theory method and means, can choose the optimal design and the measurement method of the data parsing to achieve maximum retrieve useful information from spectral data. Hyperspectral imaging can reflect the characteristics of spectrum and imaging of seed at the same time, which has been widely used in the nondestructive testing of agricultural products. In this paper, the main purpose is to study the recognition detection problem of seed varieties to find a fast non-destructive and accurately classification model with high robustness by combining the technology of hyperspectral image and chemometrics methods. The research mainly focuses on how to select effective wavelengths from hyperspectral data with chemometrics methods, and how to update model for the mathematical model established by hyperspectral image data with chemometrics methods, respectively, so as to realize the purpose of improving the purity testing accuracy of the seed. Mainly includes:1. The local learning algorithm was introduced into the optimal wavelength selection of NIR hyperspectral imaging for maize seeds. These obtained wavelengths are used to develop a discrimination model coupled with partial least squares discriminate analysis(PLSDA) to implement the rapid discrimination of maize seeds using fewer wavelengths. Hyperspectral images between 874 and 1734 nm(256 wavelengths) are acquired, using a hyperspectral imaging system, for 720 maize seed samples including 6 varieties. After that, local learning algorithm was proposed to calculate the weight values of 256 wavelengths, and the optimal wavelengths were selected according to the order of weight values. The experimental results showed that local learning algorithm can effectively select the optimal wavelengths, which could provide a suitable technical means for the rapid discrimination of maize seeds.2. The model updating based on active learning algorithm was studied using near-infrared hyperspectral imaging in the spectral region of 874-1734 nm. Active learning algorithm was investigated to add representative samples to expand the sample space for the original model, so as to implement the rapid update of the model’s ability. Random selection(RS) and Kennard- Stone algorithm(KS) were performed to compare the model update effect with active learning algorithm. The experimental results indicated that the prediction accuracy of updated model by using active learning to add a few new samples was obviously improved. Meanwhile the effect of model updated by active learning algorithm was better than that of RS and KS. Therefore, the update for purity detection model of maize seeds is feasible by active learning algorithm.3. The online model updating based on incremental support vector data description(ISVDD) algorithm was studied using visible near-infrared hyperspectral imaging technology in the spectral region of 400-1000 nm for 2000 maize seeds including 4 varieties. This study mainly aimed to update the classification model of LSSVM to improve the robustness and reliability of the model online. ISVDD was applied to determine new samples in online recognition. The new samples were added into the original hyperspace to expand the sample space of the training set. This study showed that combined hyperspectral imaging and model updating could be an effective method for classification of seeds of different years.