Assessing the Feasibility of Low-Cost Multispectral Sensing for Woody Breast Severity Classification

High equipment cost and operational complexity continue to restrict the widespread industrial adoption of spectral sensing technologies like hyperspectral imaging and near infrared spectroscopy for meat quality evaluation and wavelength selection studies show that key information can be captured using a small number of discrete bands in low-cost multispectral systems. This study evaluates the performance of the low cost SparkFun AS7265x multispectral sensor, 18 channel visible and near infrared chipset spanning 410 to 940 nm, for three class poultry meat quality classification. A total of 178 broiler breast fillets were measured at multiple anatomical regions, with repeated readings averaged to enhance signal stability. Wavelength specific statistical analysis demonstrated significant differences at 14 of 18 bands, with the strongest separations occurring in the near infrared region between 680 nm and 940 nm, confirming that discrete NIR channels drive severity discrimination. Random Forest model has achieved 92% test accuracy for three class classification. Despite lacking full hyperspectral coverage and the 980 nm water absorption maximum, the sensor captured distributed spectral signatures associated with pigment variation, fibrosis related scattering, and altered water binding. These findings demonstrate that critical quality information can be preserved within a simplified, factory calibrated multispectral architecture at a fraction of the cost of conventional systems. The results establish a foundation for translating this approach into a portable handheld device for scalable, in plant, non-destructive poultry quality assessment.