Study on hyperspectral detection and identification of invisible damage on kiwifruit by deep learning

Aiming at the problem that kiwifruit invisible damage is difficult to detect and identify by conventional detection methods, this paper proposes to use the visible near-infrared hyperspectral imaging technology to detect the identify and identify models based on deep learning VGG-16 neural network. Detection and recognition of hyperspectral images of kiwifruit invisible damage. The network is implemented by the caffe framework and python and is a 16-layer deep learning neural network. The reflection spectroscopy images of 50 kiwifruit samples were obtained at wavelengths of 400-1000 nm. According to whether they were subjected to invisible damage, they were classified into invisible damage and undamage, with 40 and 10 samples respectively. The training set and the test set are used to obtain the implicit damage discriminant model by using the principal component analysis image obtained from the spectral data as the input image of deep learning. The experimental results show that the highest accurate recognition rate reaches 100% and has a good recognition effect.