The Problem:
The quality of green coffee beans plays a critical role in the final coffee product. Contaminants such as stones, sticks, and other foreign materials are often present in raw coffee, posing challenges to both producers and consumers (Figures 1 & 2).
Manual sorting is widely used but is labor-intensive and prone to error. Short-Wave Infrared (SWIR) multispectral imaging offers a more efficient, non-destructive, and highly precise approach for detecting and removing debris from green coffee beans.
What is SWIR Multispectral Imaging?
Short-Wave Infrared (SWIR) imaging refers to wavelengths in the range of 1000 to 2500 nm, which fall just beyond the visible spectrum and into the infrared range. SWIR imaging is particularly useful for examining materials based on their moisture content, chemical composition, and structural properties. This makes the SWIR spectrum particularly well-suited for determining health and status of agricultural crops remotely. Multispectral imaging in agriculture is gaining in popularity both in the field with aerial drone reconnaissance and in the factory processing plant to identify plant blights and defects as well as detection of foreign bodies.
When applied to green coffee beans, SWIR imaging can effectively differentiate between the beans and contaminants, offering a robust tool for automated quality control. The Multispectral part comes in with each image being broken up into smaller wavelength ranges along the SWIR spectrum. For the camera used in this1 example, nine images were captured, each about 30nm wide and centered on the following wavelengths: 1070nm, 1129nm, 1188nm, 1250nm, 1311nm, 1376nm, 1436nm, 1496nm, 1553nm (figure 3). These different ranges of the SWIR spectrum can offer different information on the subjects in the image and can be combined in unique ways to highlight one feature or another.
A Simple Example
Using the Silios CICADA, a multispectral one-shot SWIR camera, a single image was captured. This camera captures nine wavelength bands simultaneously, each about 30nm wide, and provides this data as nine separate images. Combining the images to get a general SWIR image, the wood debris can be distinguished better than in the visible image, but the rocks still look like coffee beans (Figure 4).
Looking at the spectra of the different components we see that the Green coffee beans, the background and the debris all share a similar profile (Figure 5). The Green coffee beans, however, drop in intensity more than the other two between bands one (centered at 1070nm) and four (centered at 1250nm). In the longest wavelength band (centered at 1553nm) the Green coffee bean intensity drops off whereas the other two see slight increases in intensity.
By emphasizing the 1070nm band and subtracting the bands at 1250nm and 1553nm, one can create an image that enhances the intensity of Green coffee beans over the debris and the background (Figure 6). This combined image illustrates how multispectral imaging in agriculture can help with discerning desired product from debris. Armed with this image where the Green coffee beans are clearly distinguished from both the background and the three types of debris, it is easy to imagine how an automated system could remove the unwanted elements.
