Applications of Hyperspectral Imaging in Agriculture - Precision Agriculture and Fruit Sorting

Jessica Wu（China Productivity Center Agricultural Innovation Department I ）

Introduction to Hyperspectral Imaging

An optical spectrum is created by the scattering of light in an arrangement of different wavelengths. Spectral imaging works by capturing multiple optical spectrum bands simultaneously. For example, visible light is in the wavelengths of between 380nm and 780nm, ultraviolet light in the wavelengths of shorter than 380nm, infrared light in the wavelengths of longer than 780nm. With a combination of various filters and multiple photosensitive films, spectral imaging can receive different spectral bands at the same time and obtain photos in different spectral bands. Cameras are spectral imaging developed and based on visible light.

Furthermore, multispectral imaging and hyperspectral imaging are spectral imaging that extends to infrared light and ultraviolet light. Hyperspectral imaging allows point-by-point spectral analysis to detect and examine in detail the unique spectra that are imperceptible to the naked eye. It also provides information on the surface reflectance spectrum with a near-continuous distribution over large areas. This offers more dimensions of information for computation in the establishment of estimation models and highlights insightful information in practical applications. Hence, this article explores the application of hyperspectral imaging in agriculture.

By measuring the light reflected from the interaction between light and subjects such as plants, hyperspectral imaging can identify the distribution of surface vegetation, assess the condition of plants on agricultural fields, and even select fruits after harvesting. However, once the excessively repetitive and similar spectral parameters have been obtained, how to filter out distinctive wavelengths and conduct analysis becomes an important issue.

Commonly used parameter transformations nowadays include Principal Component Analysis (PCA), Band Ratio, and Normalized Difference Vegetation Index (NDVI). The transformation of machine learning parameters converts images into visualized data, so that farmers can quickly make judgments based on the image presentation at any time.

For example, the reflectance spectrum of chlorophyll is one of the important wavelengths for plants. Based on NDVI images, computers can determine the chlorophyll content and nitrogen content in plants. This allows farmers to decide on the amount of fertilizer to apply or judge whether plants are fresh (performing photosynthesis and containing moisture) or withered (not performing photosynthesis) by referring to visualized data, without setting foot on the field.

Applications of Hyperspectral Imaging in Precision Agriculture

To stay on top of the physiological condition of vegetation, farmers can mount hyperspectral instruments on drones to capture a complete picture of the farmland and observe growth factors that affecting rice, such as soil, fertilizers, and moisture by monitoring spectral parameters. In addition, the operating efficiency can be improved with drones equipped with spraying systems. This enhances the precision in the administering of water, fertilizer, and pesticides and reduce unnecessary waste.

For example, rice plants are susceptible to seedling issues, water management problems, and environmental impacts, which can lead to seedling death or lodging. It is possible to use image to zoom in on plant positions and determine the spots for replanting, thereby preventing output reduction. During the cultivation process, nitrogen fertilizer is the most important element for increasing rice yield. However, the timing and quantity of its application can affect different yield components (e.g., number of panicles, spikelets per panicle, filled-grain percentage and 1,000-grain weight).

Currently, it's also difficult to visually determine the nitrogen content in plants. Therefore, hyperspectral imaging can quickly generate a spatial distribution map of nitrogen content in rice plants on the paddy field. This allows for the development of assessment modules for yield components, nitrogen content, and subsequent yield. On this basis, the impact of nitrogen fertilizer volumes on production can be estimated. This helps avoid stem elongation and lodging due to excessive administering of nitrogen fertilizers and prevent pests, diseases or even reduced percentage of filled grain. This cuts down fertilizer expenses and mitigates environmental and ecological damage caused by heavy levels of fertilization.

Furthermore, when rice enters the milk stage and yellow ripe stage, lodging may occur due to factors such as rainy seasons, pest infestations and diseases, and this leads to significant yield reduction. In event of large-scale lodging, a damage assessment is necessary to evaluate the losses and apply for compensation. In the past, disaster appraisals were conducted manually, often resulting in labor shortages and low efficiency. The use of hyperspectral imaging can effectively enhance the efficiency and accuracy of damage assessments.

Applications of Hyperspectral Imaging in Fruit Sorting

Currently, the quality assessment of harvested fruits is based on appearance and weight, or through sampling and destructive analysis by extracting juice to measure sweetness and acidity. However, neither of these testing methods can determine whether the fruit is internally damaged. The sample results cannot represent the quality of the entire batch of fruit. This often leads to a lack of confidence among consumers in fruit quality.

Taking pineapples as an example, the most common testing method is manual tapping to distinguish whether a pineapple is flesh sound, drum-sound or wooden-pillar-hitting sound. This method is time-consuming, labor-intensive and unable to clearly determine the quality of the fruit flesh. This was the reason why Taiwan Agricultural Research Institute introduced the hyperspectral imaging technology and used machine learning to analyze spectral data that is indistinguishable to the human eye, in order to establish an effective model for determining fruit quality. Taiwan Agricultural Research Institute indicated that the current hyperspectral detection method is already rather accurate in determining the sugar content of pineapples. Once the detection model is established, sorting machines can be developed to boost the efficiency of fruit quality management in the future.

Taiwan has applied the detection technology to pear selection to create “Lightwave Asian Pear” brand. This technology employs near-infrared light to determine the sugar content and acidity of pears, in conjunction with appearance screening, grading and packaging. The purpose is to select high-quality pears and enhance consumers’ confidence in the brand.

Conclusion

In face of climate change and labor shortages, smart agriculture is becoming the future trend. The deployment of hyperspectral imaging allows farmers to better understand the field conditions. In addition to output prediction, it can also detect pest infestations and incubation periods of diseases, assess the status of plant growth and reduce the required volume of fertilizers and pesticides. In fact, the practice of precision agriculture also improves food safety and quality. Currently, hyperspectral imaging is not only used in agriculture but also applied in food science, medical care, and criminalistics. Perhaps in the future, hyperspectral imaging will be adopted more widely in assessment models and applications.