Identification of Bruised Apples Using Deep Learning and 3D Near-infrared Imaging

In this paper, we propose an algorithm for recognizing bruised apples based only on surface shape information obtained by a 3D near infrared (NIR) imaging system. The proposed algorithm is composed of two parts: construction of feature map, and classification of apples into bruised or unbruised categories. We propose a new algorithm to code 3D shape information into a 2D feature map. For classification, we propose to build a convolutional neural network to extract deep hierarchical features from the 2D feature maps that are optimal for the identification of bruised apples. Experimental results show that the proposed algorithm is better than the algorithm developed previously, which indicates the potential of the proposed algorithm for the identification of bruised apples.