Monitoring and controlling food quality is of the most significant interest for profitable and responsible food production. Especially fruits and vegetables are more sensitive than other food products and must be sold and processed fresh to be valuable and healthy. Hyperspectral imaging provides crucial data for automatic quality control systems to assure the high quality of food products.
Measuring the aging of plums and tomatoes with Specim FX10 hyperspectral camera
Food aging is an important parameter to be quantified when evaluating freshness. Within this context, the ripeness and the firmness of fruits and vegetables are the two most essential qualities to be observed and monitored. Hyperspectral cameras enable observation of spectral changes in fruits and vegetables throughout the ripening process.
In this study, we used the Specim FX10 hyperspectral camera and a lab scanner to inspect plums and tomatoes for 20 days to assess the aging process (Figure 1). The Specim FX10 is a visible-near infrared (VNIR) camera that covers the spectral range from 400 to 1000 nanometers. The first part of the analysis focuses on the spectral features of the samples over time. Then, a regression model of tomatoes’ and plums’ aging is presented.
Figure 1: A sample of three plums and tomatoes placed on a lab scanner 40×20 and measured with a Specim FX10 camera for 20 days.
Photos of the samples were taken, along with the hyperspectral data. The pictures show that the freshness of the plums, especially the tomatoes, degraded firmly over time (Figure 2). A small cut was made in the middle of one tomato and plum. It seemed to have a substantial impact on accelerating the aging of the tomato but not on the plum.
Figure 2: Photos of the samples taken on the 1st, 13th, and 20th day.
Spectral reflectance reveal chemical changes
A rectangular selection was made on each plum and tomato each day when the spectral measurements were made (1st, 2nd, 3rd, 6th, 9th, 13th, 14th, 16th, 17th, and 20th day). Only the spectra obtained on the 1st, 13th, and 20th day are presented in Figure 3 to ease the reading of the results. Spectra are averaged over the selection.
The spectral differences are more significant for the tomatoes than for the plums. This is already visible in the photos taken on the 1st, 13th, and 20th days (Figure 2).
The spectra reveal chemical changes which happen over time within the fruits and vegetables. Plums and tomatoes are green at early growing stages due to the chlorophyll they contain. But when ripening, the chlorophyll breaks down into another chemical. For tomatoes, chlorophyll breaks down into lycopene, which explains the red color. This chemical change explains the spectral variation of the plums and tomatoes over time between 550 and 750 nanometers. The ripening process of the fruits and vegetables also affects the moisture level or structure, impacting their spectra at 970 nanometers. Other properties (e.g., sugar content) also change over time, shaping the spectral reflectance.
Figure 3: False RGB images of the plum and tomatoes acquired on the 1st, 2nd, 3rd, 6th, 9th, 13th, 14th, 16th, 17th, and 20th day. Each dataset was combined into a single one (mosaic), from the left (1st day) to the right (20th day). Averaged spectra for each tomato and plum are displayed on the 1st day (white), 13th day (pink), and 20th day (purple).
Regression model to quantify the aging
A regression model was built to quantify the aging of the plums and tomatoes (Figure 3). The imaging day was the actual regression variable.
With the plums, the R2 was 0.81, whereas, for the tomatoes, it was 0.91. Those were computed on other selections than those used to train the model. The regression graph of Actual value vs. Predictions is presented in Figure 4.
For the plums, the model was based on the reduced spectral range from 588 to 976 nanometers. For the tomatoes, the model was based on the spectral bands between 445 and 993 nanometers.
Figure 4: Regression model output on the three plums (top) and three tomatoes (bottom). Data were acquired on the 1st, 2nd, 3rd, 6th, 9th, 13th, 14th, 16th, 17th, and 20th day (from left to right). The heat map ranges from the 1st day (Min) to 25th day (Max).
Figure 5: Actual values vs model predictions for both models (measuring the aging of plums and tomatoes).
Conclusion
The Specim FX10 camera is suitable for measuring fruits and vegetables’ ripeness and aging as it is sensitive to traits related to freshness for agri-food products. When building a typical regression model, laboratory measurements should be used as a reference value to develop and validate the model. However, those are not needed for accessing fruits and vegetables’ aging.
Hyperspectral cameras operating in visible-near infrared (VNIR) provide an efficient tool for monitoring the product quality of fresh food products. Hyperspectral imaging is an especially suitable method for food grading, sorting, and classification compared to conventional point-based methods due to its non-destructive nature.
