Auto-encoder design based on the 1D-VD-CNN model for the detection of honeysuckle from unknown origin

The disadvantages of the traditional one-dimensional convolution neural network (1D-CNN) model based on honeysuckle near-infrared spectral data (NIRS) include high parameter quantity, low efficiency, and inability to identify unknown categories effectively. In this paper, we propose a one-dimensional very deep convolution neural network (1D-VD-CNN) and design an auto-encoder mechanism for detecting honeysuckle from unexplored habitats. First, the 1D-VD-CNN model uses the efficient very deep (VD) structure to replace the hidden layer structure in the traditional 1D-CNN model. The model can be directly applied to analyze one-dimensional near-infrared spectral data (NIRS). Second, combining the reconstruction error of the auto-encoder, a honeysuckle identification method considering an unknown origin is designed, which can solve the problem of high confidence in convolution neural networks by using an auto-encoder and reconstruction errors of the samples to be tested. Whether the sample is an unknown variety can be determined by comparing the corrected confidence level with the preset threshold value. The results show that the accuracy of the 1D-VD-CNN training set and test set is 100%, and the loss value converges to 0.001. Compared with the traditional 1D-CNN model, the parameters and FLOPs are reduced by nearly 71% and 8%, respectively. At the same time, compared with the NIRS analysis and the PLS-DA method, the 1D-VD-CNN model has higher efficiency and better recognition performance for honeysuckle near-infrared spectral classification. Meanwhile, the accuracy rate of the auto-encoder for the category detection mechanism of honeysuckle from an unknown origin is 98%. The model can quickly and efficiently classify honeysuckle from different habitats and detect honeysuckle from unexplored habitats.