Quantitative Regression Modeling of Cocoa Bean Content Based on Gated Dilated Convolution Network

By analyzing the near-infrared spectrum, we can determine the quantitative relationship model between the spectral data of different cocoa beans and the target components. This paper proposes a predictive regression model based on 1D-CNN. Based on the traditional convolutional neural network, gating mechanisms and dilated convolutions are combined. The particle swarm optimization method is used to optimize the hyper-parameters of one-dimensional convolution. The end-to-end near-infrared predictive regression model does not require wavelength selection. It is convenient to use and has a strong promotional value. Taking the public cocoa beans near-infrared data set as an example, the method can predict the water and fat content in cocoa beans, and the effectiveness of the method is verified. Comparing the improved one-dimensional convolution with traditional one-dimensional convolution results and partial least squares regression, it shows better prediction accuracy and robustness.