Reliable image processing algorithm for sunburn management in green apples

To tackle weather uncertainties and associated heat stress to apple fruits, researchers have been exploring development of a real-time crop stress monitoring systems. Our group have been researching one such in-field sensing system that uses localized weather and thermal-RGB imagery proceed on the edge for monitoring fruit surface temperature (FST). Such solutions can be tied with mitigation techniques (e.g., water-based cooling methods) as precision management. However, current edge compute algorithms are limited to segment thermal-RGB imagery for the red pigmented fruits near maturity and lack the green fruit segmentation, limiting the usability of the in-field sensing system. Aim of this study was to develop and validate a color independent fruit segmentation algorithm for successful FST monitoring. Longwave infrared (LWIR) imagery at daily peak air temperature was utilized to achieve temperature gradient aided fruit segmentation and to estimate FST for next 24-h. The algorithm robustness was field evaluated in Fog-Net (combination of fogging and netting) cooling and control treatments (Year 2021). Overall, algorithm accurately detected fruits in early growing season when fruits are green and effectively captured the treatment effects based on FST data. Additionally, the algorithm was also evaluated for computational overhead and estimated FST accuracy on an in-field sensing node (Control treatment) deployed in commercial apple orchard (Year 2022). CPSS took 12 milliseconds to process LWIR image with no CPU throttles and there was no significant difference between $\boldsymbol{(\mathrm{R}^{2}=0.98}$, p-value = 0.1607) FST estimated using LWIR (FSTi) and thermal-RGB based (FST_Actual) image processing algorithm.