Wenchuan Guo

College of Mechanical and Electronic Engineering, Northwest A&F University,
Yangling, Xi'an, China

China’s fruit production accounts for 30% in the world, and its output has increased continuously in the past decade. Since 2019, the production value of fruit in China is about 400 billion RMB. Sugar content and firmness are the key factors that determine consumers’ desire to buy fruit and the commercial value of fruit. However, the existing fruit sugar content detector developed based on visible/near-infrared spectrometer has the disadvantages of high price and high power consumption. Compared with visible/near-infrared spectroscopy technology, multi-spectral technology makes the development of fruit quality detector with low cost and low power consumption possible. However, at present, most quality detectors based on multi-spectral technology can only detect sugar content of a single variety of fruit. How to break through the problem of high efficiency and precision, and realize the internal quality detection of many kinds of fruit is the key to develop the detector based on multi-spectral technology.

To overcome these problems, a probe for detecting the internal quality (sugar content and firmness) of different kinds of fruit (kiwifruit, pear and apple) was developed based on the optical simulation. Then a handheld multi-fruit internal quality nondestructive rapid detector, which can realize the non-destructive and simultaneous detection of sugar content and firmness, was developed. To improve the detection accuracy, a high-quality pseudo-full spectrum generation method based on deep convolutional generation adversarial network (DCGAN) was proposed. This method used the multi-spectra as input to generate pseudo-full spectra with high similarity to the "real" full spectra, taking full advantage of DCGAN’s powerful feature extraction ability and excellent sample generation quality. With pseudo full spectra as the input variables, the convolutional neural network model for the prediction of fruit internal quality was established. Then the pseudo full spectra generation model and fruit internal quality prediction model were realized in the mobile terminal. Finally, a fruit internal quality detection APP through bluetooth communication were designed, realizing non-destructive and high precise detection on sugar content and firmness of of different varieties of fruit. The development of this instrument is of great significance for guiding the harvesting, storage, processing, transportation and traceability of fruit.