Thông tin chung
  English
  Đề tài NC khoa học
  Bài báo, báo cáo khoa học
  Hướng dẫn Sau đại học
  Sách và giáo trình
  Các học phần và môn giảng dạy
  Giải thưởng khoa học, Phát minh, sáng chế
  Khen thưởng
  Thông tin khác
  Tài liệu tham khảo
  Hiệu chỉnh
 
Số người truy cập: 107,328,368

 Fruit recognition based on near-infrared spectroscopy using deep neural networks.
Tác giả hoặc Nhóm tác giả: Duy Khanh Ninh, Thi-Ngoc-Canh Doan, Thanh Xuan Nguyen-Thi, Nhan Le Thanh
Nơi đăng: ICMLSC '21: 2021 The 5th International Conference on Machine Learning and Soft Computing; Số: 978-1-4503-8761-3;Từ->đến trang: 90–95;Năm: 2021
Lĩnh vực: Khoa học; Loại: Bài báo khoa học; Thể loại: Quốc tế
TÓM TẮT
ABSTRACT
Near-infrared (NIR) spectroscopy has been widely used to determine the varieties and chemical properties of agricultural and food products. The major advantage of NIR spectroscopy is that the analysis is carried out in a simple, fast, and non-destructive manner, making it suitable for food applications. As the first step in applying NIR spectroscopy for fruit recognition and analysis in Vietnam, this paper presents deep neural networks (DNNs) based solutions for automatic recognition of several kinds of fruits. We compared two proposed DNN architectures based on Convolutional Neural Network (CNN) and Residual Network (ResNet). Additionally, we proposed feature extraction methods using the first and second derivatives of the Savitzky-Golay (SG) filtered normalized NIR data. Experimental results show that the deep learning approach combined with reasonable feature extraction process can achive the accuracy of approximately 99% for the task of classifying five types of fruits including Apple, Avocado, Dragon Fruit, Guava, and Mango. The ResNet-based model is more compact and has slightly better recognition performance than the CNN-based one. The inclusion of the first and second derivatives of SG-smoothed normalized spectra improves the recognition accuracy of the proposed DNN models by more than 8%. Moreover, the recognition performance of the proposed DNN models surpasses that of traditional classifiers, including k-nearest neighbors, Naive Bayes, and support vector machine. Our proposed methods were proved to be robust against the freshness of fruits, the NIR device's calibration parameters, and the measurement position on the body of fruits.
© Đại học Đà Nẵng
 
 
Địa chỉ: 41 Lê Duẩn Thành phố Đà Nẵng
Điện thoại: (84) 0236 3822 041 ; Email: dhdn@ac.udn.vn