Deep Feature Representation with Stacked Sparse Auto-Encoder and Convolutional Neural Network for Hyperspectral Imaging-Based Detection of Cucumber Defects

Abstract

Abstract. It is challenging to achieve rapid and accurate processing of large amounts of hyperspectral image data. This research was aimed to develop a novel classification method by employing deep feature representation with the stacked sparse auto-encoder (SSAE) and the SSAE combined with convolutional neural network (CNN-SSAE) learning for hyperspectral imaging-based defect detection of pickling cucumbers. Hyperspectral images of normal and defective pickling cucumbers were acquired using a hyperspectral imaging system running at two conveyor speeds of 85 and 165 mm s-1. An SSAE model was developed to learn the feature representation from the preprocessed data and to perform five-class (normal, watery, split/hollow, shrivel, and surface defect) classification. To deal with a more complicated task for different types of surface defects (i.e., dirt/sand and gouge/rot classes) in six-class classification, a CNN-SSAE system was developed. The results showed that the CNN-SSAE system improved the classification performance, compared with the SSAE, with overall accuracies of 91.1% and 88.3% for six-class classification at the two conveyor speeds. Additionally, the average running time of the CNN-SSAE system for each sample was less than 14 ms, showing considerable potential for application in an automated on-line inspection system for cucumber sorting and grading. Keywords: Convolutional neural network, Defect detection, Hyperspectral imaging, Pickling cucumber, Representation learning, Stacked sparse auto-encoder.