Near-infrared spectroscopy has become an indispensable analysis method in industrial and agricultural production process quality monitoring. It has been widely used in the qualitative and quantitative analysis of food, agriculture, medicine and others.-A near-infrared spectroscopy prediction model with high prediction accuracy, high-speed running,and strong generalization ability plays an essential role in the qualitative and quantitative analysis of different substances. However, due to the increase innear-infrared spectroscopy data, the disadvantages of traditional near-infrared spectroscopy modeling methods are obvious. With the development of artificial intelligence technology, deep learning algorithms have been widely used in the field of near-infrared spectroscopy. The quantitative analysis model of near-infrared spectroscopy based on a parallel convolution neural network (PaBATunNet) was proposed. PaBATunNet comprisedone1-D convolutional layer, one parallel convolution module (Module), one flattening layer, four fully connected layers and one parameter regulator (PR).The Module included five submodules and one Concatenate function, which was used to extract the linear and nonlinear multidimensional features of the spectral data, respectively and concatenate them. The prediction accuracy of PaBATunNet was improved by PR, which optimized the model parameters. The high contribution characteristic wavelengths of PaBATunNet were given based on Gard-CAM, which improved the interpretability of PaBATunNet. By taking public near-infrared spectroscopy datasets of grain, diesel fuel, beer and milk as examples, the prediction results of PaBATunNet were compared with partial least squares (PLS),principal component regression (PCR), support vector machine (SVM) and back propagation neural network (BP). The results showed that the prediction accuracies of PaBATunNet tograin, diesel fuel, beer and milk datasets were respectively increased by 30.0%, 40.7%, 43.0% and 52.8% in comparison with PLS, 28.8%, 35.9%, 40.8% and 52.2% in comparison with PCR, 45.5%, 37.4%, 45.3% and 54.7% in comparison with SVM, and 7.9%, 32.4%, 90.1% and 62.0% in comparison with BP. Compared with the traditional near-infrared spectroscopy modeling methods, the PaBATunNet based on the parallel convolutional neural network, has solved the problems of low prediction accuracy, long running time, poor generalization ability and poor interpretability. It can be effectively applied to quantitative analysis in industrial and agricultural production. It provides a theoretical basis for establishing the rapid, nondestructive and high-precision near-infrared spectroscopy quantitative analysis model.