Nondestructive perception of postharvest damage and storage deterioration analysis of apples based on multi-source sensing data and machine learning

Apples are prone to mechanical damage from drops during postharvest handling, reducing commercial value. This study on “Fuji” apples examined the effects of drop heights (0–120 cm) on storage quality and dielectric properties, and built a machine learning damage detection model using multi-sensor data. Over 7 days, quality indicators (firmness, soluble solids, weight loss, colour, damaged area, sensory scores), dielectric parameters, and microenvironment sensor data (CO 2 , VOCs) were measured. Results showed firmness decreased and weight loss increased with height; at 120 cm, firmness dropped 24.96%, weight loss reached 6.15%, sensory score fell to 0.6, and damaged area expanded rapidly. Dielectric parameters showed stable frequency responses at 100 kHz, distinguishing damage degrees. CO 2 and VOCs rose with height, with CO 2 2.09 times that of the control at 120 cm. Among the four models (SVM, ELM, RBF network, and BPNN), BPNN achieved 88.96% test-set accuracy and a Kappa coefficient of 86.75%. The sensing system achieved a 100% data upload success rate during a 48-day unattended operational test. This provides methodological support for non-destructive postharvest apple damage detection.