A novel and interesting experiment based on the association of artificial intelligence and visible near-infrared spectroscopy (Vis-NIRS) techniques was conducted to rapidly perceiving the living habitat of rice in China.Three distinct classes of rice collected from the Hainan, Jiangsu and Heilongjiang provinces in China which respectively corresponded to the megathermal,mesothermal and microthermal climate living habitats were investigated.The field spectroradiometer (FieldSpec 3 FR, Analytical Spectral Devices (ASD), Inc.USA) with wavelength range between 325 and 1075 nm with high resolution of 1 nm was used to collect the reflectance Vis-NIRS signals of rice.The raw spectral data was preprocessed by using the multiplicative scatter correction algorithm in order to remove scaling, path length, baseline offset etc.effects.The swarm artificial intelligence method of ant colony optimization algorithm (ACO) was adopted for searching the fingerprint wavelengths from the raw spectrum.A total of 18 sensitive feature variables were attained by the metaheuristie algorithm.The principal component analysis algorithm was used to extract the principal components from these optimum characteristics.Three kinds of recognition methods including partial least square regression (PLSR) algorithm, back-propagation artificial neural network (BP-ANN) perceptron and multi-class support vector machine (MCSVM) were performed to correlate the logical living habitat attributions with the attained principal components of rice Vis-NIRS.The prediction performance of each model was optimized by using the leave-one-out cross-validation approaches.Each type of the training set and test set contained 100 and 20 instances, respectively.The dataset of rice Vis-NIRS was randomly divided into two parts for the training and test purpose.The perceiving precision computed by the PLSR algorithm only reached to 95%.It is because that the measured reflectance spectrum of rice displayed the strong nonlinearity, the linear modeling method of PLSR have difficult to discriminate the logical living habitat attributions in the traditional linear space,which led to the accuracy decline.The radial basis function (RBF) was chosen as the transfer function of BP-ANN model.The RBF could be used to alleviate the effects of nonlinearity of the measured reflectance spectrum to some extent.The perceiving precision computed by the BP-ANN algorithm rose to 97%.Compared with the prediction results of three proposed models, the prediction performance of MCSVM was superior to the other two, whose final recognition accuracy achieved 100%, so the relationship between the nonlinear Vis-NIRS data and the logical living habitat attributions of rice was fully disclosed.It is because that the kernel-based forecast model of MCSVM could eliminate the impact of nonlinear Vis-NIRS data through mapping the original nonlinear spectral data to the high dimensional linear feature space.The results indicated that the chosen characteristic wavelengths contributed to optimizing the performance of the perceiving models and the nonlinearity of the measured reflectance spectrum of rice could be well correlated with the logical living habitat attributions by using kernel-based forecast algorithm of MCSVM.It could be concluded that the explored approaches of combination of swarm artificial intelligence methods of ACO and MCSVM with Vis-NIRS techniques could be successfully utilized to rapidly and accurately perceive the living habitats of rice in China.