Research progress in hyperspectral imaging technology for fruit quality detection

Abstract: Fruit has rich nutritional value, which occupies a large proportion in human production and life. It can not only help fruit growers to gain economic value, but also provide all kinds of nutrients needed by people of all ages. According to the World Food and Agriculture Organization (WFAO), the global annual fruit production is about 800 million tons, and fruit consumption is also increasing year by year. With increasing the economic value of the industry, fruit sales also consider how to meet cus- tomers’demand for fruit quality. The fruit quality has become the common concern from fruit suppliers and consumers. At present, fruit quality inspection by artificial vision is still widely used, but it is a sub- jective, time-consuming, laborious, cumbersome and inaccurate method. The commonly used instru- ment detection means are mainly analytical and chemical methods, like mass spectrometry and high per- formance liquid chromatography. However, they also have many limitations, for instance, being destruc- tive, time-consuming and unable to process large numbers of samples, and require large amounts of time to prepare samples. Therefore, it is vital and necessary to apply accurate, reliable, efficient and non- destructive alternative methods to evaluate fruit quality and other quality-related attributes. Hyperspec- tral image (HSI) can provide spatial and spectral information, in the continuous range of wavelengths to produce a series of high resolution image information, and the data with one dimension spectral infor- mation and two dimensions spatial information, can constitute the three-dimension hyperspectral cube, and therefore, each pixel of hyperspectral image can save the corresponding position of the spectral in- formation. The obtained spectrum has the function of reflecting the information about this particular pix- el. Hyperspectral image can quickly and nondestructively detect the physical and morphological charac- teristics of fruits as well as the inherent chemical and molecular information, and is becoming a power- ful analytical tool for fruit quality detection. This paper reviews the progress and application in hyper- spectral imaging for fruit quality evaluation in the last ten years, and the latest progress and application in hyperspectral imaging system for the detection, classification and visualization of fruit quality and safety attributes are introduced. The basic principle and main instrument composition of hyperspectral imaging system are introduced. The methods of hyperspectral data acquisition, preprocessing and mod- eling are summarized. In addition, the methods for measuring the external and internal characteristics of fruit, such as firmness, titratable acidity (TA), soluble solids content (SSC) and moisture content (MC) in the last ten years are also discussed and tabulated. The fruit real-time monitoring system based on hy- perspectral imaging technology is expected to meet the requirements of modern industrial control and sorting system in the near future and provide reference for the intellectualization of fruit industry. The research progress in hyperspectral imaging technology for fruit quality detection is as follows: (1) Fruit scratch detection based on hyperspectral imaging technology mainly focuses on apple, kiwifruit, straw- berry, jujube and other fruits, among which apple scratch detection is the most popular. The introduc- tion of hyperspectral imaging technology improves the prediction efficiency of fruit bruising to distin- guish bruising from normal fruits and bruising with different depth. However, image processing tech- niques should be used with caution when using hyperspectral techniques to detect minor abrasions. In addition, the above research is limited to a few varieties of a certain type of fruit, so it is necessary to further study more fruit materials of different varieties. (2) The hyperspectral imaging system has been successfully applied to the chilling injury identification of apples, peaches and jujubes, but there are few literatures on the chilling injury identification of tropical and subtropical fruits that are more suscep- tible to chilling injury. (3) The spatial resolution technique of hyperspectral imaging has a very wide range of evaluation on fruit hardness, which has been used to measure the hardness of most fruits, such as apples, peaches, bananas, pears, cherries, persimmons, plums, mangoes, blueberries, etc. In addition, the classification of fruit maturity based on hardness also shows great potential, but it needs to be fur- ther improved by increasing population size, secondary sampling method and improving measurement conditions. (4) Hyperspectral absorption imaging technology can be used to evaluate the content of solu- ble solids in apples. However, the current research is limited to the equatorial position of the apple. In order to obtain more reliable and comprehensive prediction results, different algorithms need to be used to deal with more different positions of the apple. Future research will focus on the surface distribution of solid materials based on hyperspectral reflection imaging systems. In addition, in order to improve the calculation speed and modeling accuracy, it is necessary to reduce the high dimension of hyperspec- tral data. Future research should explore further data mining to reduce redundant hyperspectral data without losing valuable information, extend the feasibility of new algorithms to develop stable predic- tive models, and improve the accuracy of the models. (5) The hyperspectral imaging technology to pre- dict fruit TA content is rarely used, mainly including oranges and mangoes, and TA is more used to pre- dict fruit ripening time. (6) The accurate prediction of fruit moisture content by hyperspectral imaging technology has not been satisfactory, and it still needs to be further explored.