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.

Nondestructive Detection of Moldy Core in Apples Based on One-dimensional Convolutional Neural Networks

Published by the American Society of Agricultural and Biological Engineers, St. Joseph, Michigan www.asabe.org

Citation:  2021 ASABE Annual International Virtual Meeting  2100328.(doi:10.13031/aim.202100328)
Authors:   Zhongxiong Zhang, Haoling Liu, Bin Li, Yuge Pu, Juan Zhao, Jin Hu
Keywords:   Apple; Moldy core; VIS/NIR Spectroscopy; 1D-CNN;

Abstract. The realization of rapid non-destructive testing of moldy core in apple is of profound significance to protect consumer health and property safety. The moldy apple core discriminant model established by traditional machine learning methods has the problem of low accuracy. In this paper, the one-dimensional convolutional neural network (1D-CNN) based model for nondestructive detection of moldy apple core was proposed. Firstly, the apple transmission detection system with a spectral range of 200-1100nm independently built-in by the laboratory was used to obtain the visible/near-infrared transmission spectral information of the samples. The relationship between the topology of the convolutional neural network, the parameters of the convolutional kernel function (size, number) on the model is discussed separately. Principal component analysis (PCA) was utilized to reduce the dimensionality of the spectral data, and the first nine principal components that could represent 99.9% of the original spectral information were selected as the input data for the conventional model. Finally, the 1D-CNN was compared with traditional machine learning models partial least squares linear discriminant analysis (PLS-DA) models and support vector machine (SVM) nonlinear discriminant models, the discriminant accuracy of PCA-PLS-DA, PCA-SVM, and 1D-CNN models were 89.84%, 93.55%, and 98.39%, respectively. Compared with the traditional model, the proposed model based on1D-CNN has higher recognition accuracy. The results show that the deep learning method has significant research value and broad application potential in the field of fruit nondestructive testing.

(Download PDF)    (Export to EndNotes)