A Multispectral Imaging System to Assess Meat Quality

Multispectral imaging uses reflectance information of a number of discrete spectral bands to classify samples according to their quality defined using standard parameters. A multispectral image is rich in information compared to a normal RGB image. Therefore, a multispectral image can be used to classify samples more accurately than an RGB image. This paper discusses a design of a multispectral imaging system that can be used to assess the quality of meat. The system is comprised of six LEDs with nominal wavelengths between 405 nm and 740 nm. The light emitted from LEDs reach the meat sample placed inside a dark chamber through an integrating hemisphere. LEDs are lighted one at a time and images of the meat sample are captured for each flash separately using a smartphone camera. Eventually, all the images of the meat sample, taken at a specific time instance were integrated to form the multispectral image. The meat samples stored at $4 \circ \mathrm {c}$ were imaged up to four days at predetermined time intervals using the designed system. Once the data acquisition was completed, all the pixels of the multispectral image were represented as points in high dimensional space, which was then reduced to a lower dimensional space using Principal Component Analysis (PCA). It was observed that images of meat sample obtained at different time instances clustered into different regions in the lower dimensional space. The experiment was performed with chicken meat samples. This proves the viability of using multispectral imaging as a non-invasive and non-destructive method of assessing meat quality according to certain quality parameters. Off-the-shelf electronic components and a regular smartphone were used to build the system, thus making the system cost-effective.