Explainable Droplet Recognition System for Precision Sprayer Applications

AI-driven detection systems are playing an increasingly important role in the advancement of precision agriculture. In this paper, we have implemented a transfer learning pipeline for water droplet detection with the intent to develop quantifiable and real-time detection of post-spray areas for precision spraying applications. The object detection pipeline effectively identified multiple features for water droplet detection from the three curated datasets. We have used two pre-trained convolutional backbones as the feature extractor and achieved an overall detection mean average precision across the three curated datasets of 0.409 and 0.277 for the ResNet50, and MobileNetV3-Large backbones respectively. Additionally, for visual explanations and interpretation, we implemented EigenCAM class activation mapping techniques to highlight the regions of the input images that are important for predictions.