Internal browning detection in red-flesh apple (<i>Malus</i> <i>domestica</i>) using image analysis and acoustic signal-based detection

Uncorrelating red-flesh color and internal browning is a breeding target in the selection of red-flesh elite apple cultivars, and accurate phenotyping methods are needed to study the genetic architecture associated with internal browning. Image analysis combined with machine learning and two-stage data acquisition enabled the detection of seven QTLs for brown-related traits on LG1, LG9, and LG10 in four interconnected red-flesh F1 families. A major QTL on LG10, which colocalized with a polyphenol oxidase (PPO), was linked to early internal browning susceptibility. However, internal browning is a temporal and spatial process that involves biochemical (PPO-mediated phenolic oxidation) and structural (cellular integrity) factors. Non-destructive and accurate methods are required to evaluate internal browning throughout fruit senescence. For this purpose, acoustic signal-based detection of internal browning in red-flesh apple was investigated in one red-flesh parent. Machine learning models showed good accuracy in discriminating 'brown' and 'non-brown' signals. cGAN data augmentation outperformed model performances suggesting that extended datasets should lead to better accuracy in internal browning detection. This study shows that acoustic measurement could be a valuable non-destructive tool to discriminate between brown and non-brown apples with potential applications for phenomic selection and/or automatic apple sorting. Altogether, this study provides insights into internal browning physiology in red-flesh apple with future applications in red-flesh breeding.