Click on “Download PDF” for the PDF version or on the title for the HTML version.

If you are not an ASABE member or if your employer has not arranged for access to the full-text, Click here for options.

Abstract. Data preprocessing is a challenging task for near infrared (NIR) spectroscopic model with conventional chemometrics techniques. Improper use of preprocessing may not reduce experimental and instrumental artifacts, but introduce manual artifacts and result in worse performance. Developing an end-to-end data-based learning approach without preprocessing can possibly improve the model performance, reducing the requirement of prior knowledge and human efforts. A deep learning approach based on convolutional neural network (CNN) is developed for classification analysis of near infrared spectral data. The model has two convolutional layers and two fully connected layers. We compare the CNN model to partial least squares regression – linear discriminant analysis (PLS-LDA) and principal component analysis - logistic regression (PCA-LR) models on a NIR dataset for grapevine classification. The results show that the CNN model is able to learn the insight patterns directly from the raw data, which achieves a higher classification accuracy than PLS-LDA and PCA-LR methods with and without preprocessing techniques. The CNN model is more stable and less affected by dataset changes than conventional methods.