Reliability Scores from Saliency Map Clusters for Improved Image-based Harvest-Readiness Prediction in Cauliflower

Cauliflower is a hand-harvested crop that must fulfill high-quality standards in sales making the timing of harvest important. However, accurately determining harvest-readiness can be challenging due to cauliflower head being covered by its canopy. While deep learning enables automated estimation, errors occur field-variability and limited training data. In this paper, we analyze reliability classifier with interpretable machine learning. By identifying clusters saliency maps, derive scores for each classification result using knowledge about domain image properties. For unseen data, used (i) inform farmers improve their decision-making (ii) increase model prediction accuracy. Using RGB images single plants at different developmental stages from GrowliFlower dataset [4], investigate various mapping approaches find they quality scores. With most suitable interpretation tool, adjust achieve 15.72% improvement overall accuracy 88.14% 15.44% average class 88.52% dataset.