Convolutional neural network (CNN) based on hyperspectral technique for the identification of covered raspberry fruits

In this study, we developed a novel recognition method based on hyperspectral technology and a one-dimensional convolutional neural network (1D CNN), aiming to address the shortcomings of traditional machine vision techniques in recognising raspberry fruits that are completely covered by leaves. Taking advantage of the ability of hyperspectral technology to distinguish fruits behind leaves, we designed the SpectralCRF-CNN (Spectral-Covered Raspberry Fruits CNN) model specifically for spectral data, which significantly improved the recognition accuracy of covered raspberry fruits. The experimental results show that the SpectralCRF-CNN model achieves an F1 score of 0.63 on the full-band light intensity dataset, and especially performs well on Class 0 (no-covered fruit case), with an accuracy close to 0.9 and an F1 score as high as 0.82. This study provides an effective technical means for the accurate recognition of raspberry fruits, which is of great practical automation and precision farming This study provides an effective technical means for the accurate identification of raspberry fruits, which has important practical value for agricultural automation and precision agriculture.