Citrus chilling injury detection Online fruit quality inspection Non-destructive fruit testing

2025-11-04

Citrus fruits are susceptible to chilling injury during post-harvest storage and transportation. At the early stage, such damage shows no obvious external signs, yet it severely affects taste and market value. Optosky ATP2000P spectrometer, combined with an online transmission spectroscopy system, Diameter Correction Method (DCM), and deep learning algorithms, enables high-precision, online, and non-destructive detection of early chilling injury in citrus, offering reliable technical support for intelligent fruit sorting.

01 Application Background

Industry Challenges in Chilling Injury Detection

Citrus is a major global economic crop but is prone to chilling injury under low-temperature conditions. In the early stage, internal ice crystal formation causes cell dehydration and nutrient loss, while leaving no visible external traces. Traditional manual inspection or flotation methods are inefficient, destructive, and inadequate for modern sorting requirements.

02 ATP2000P Online Detection System

Fig. 1. Schematic of online transmission spectra measurement system

System Components

Light Source: 100W halogen lamp with focusing lens

Spectrometer: ATP2000P spectrometer (wavelength range: 644–1100 nm)

Transmission Mode: Online transmission for high stability

Auxiliary Equipment: Encoder trigger, shielding cover to prevent interference, automatic data saving

Diameter Correction Method (DCM)

Fruit size variation significantly affects transmission spectral intensity. Using the ATP2000P online transmission system, the innovative Diameter Correction Method (DCM) identifies the wavelength most correlated with fruit diameter (825 nm) and standardizes the absorption spectra accordingly. This effectively eliminates the influence of size, outperforming traditional preprocessing methods such as MSC and SNV.

Intelligent Modeling & Deep Learning

Traditional Models: Algorithms including PLSDA and SVM, combined with effective wavelength selection via CARS and SPA, to establish lightweight, high-accuracy models.
1D-CNN Deep Learning Model: A three-layer convolutional structure extracts deep spectral features. Data augmentation techniques enhance model generalization. The final model achieved recall rates of 95.15% for injured fruits and 88.54% for healthy fruits in the prediction set, with an overall accuracy of 91.96%.

03 Application Results & Advantages

Fig.2. Apparent characteristics of frezedamaged orange (a) and sound unfrozen orange (b),.(For interpretation of the references to colour in this figure legend, the reader is referred to the web version of this article.)

Fig.3. Absorbance spectra of all the samples acquired by the online mea-surement system.

Item	Description
Detection Target	Early chilling injury in citrus (Orah variety)
Detection Method	Online transmission spectroscopy, non-destructive
Core Algorithms	DCM + PLSDA/SVM/1D-CNN
Optimal Model	DCM-1D-CNN
Accuracy	91.96% (prediction set)
Key features	Resistant to size interference, suitable for high-speed online sorting

04 Conclusion

ATP2000P optical fiber spectrometer, integrated with diameter correction and deep learning technologies, enables online non-destructive detection of early chilling injury in citrus fruits. This system is not only suitable for citrus but can also be extended to internal quality inspection of fruits such as apples and pears, providing a comprehensive and efficient solution for intelligent post-harvest handling of fruits and vegetables.

Reference

Shijie Tian, Shuai Wang, Huirong Xu, Early detection of freezing damage in oranges by online Vis/NIR transmission coupled with diameter correction method and deep 1D-CNN, Computers and Electronics in Agriculture, Volume 193, 2022, 106638, ISSN 0168-1699, //doi.org/10.1016/j.compag.2021.106638.