Deep Learning for Precision Agriculture: Post-Spraying Evaluation and Deposition Estimation

XAI improves agricultural decision-making: The paper demonstrates that explainable AI techniques (specifically AblationCAM and ScoreCAM) can make AI systems more interpretable for precision agriculture, allowing farmers and agronomists to understand why the system made specific decisions about spray coverage rather than treating it as a "black box."

Automation reduces labor and improves consistency: The proposed computer vision pipeline automates post-spraying evaluation without requiring traditional manual agricultural assessment methods, potentially reducing the time and subjective judgment needed to evaluate whether herbicide or pesticide applications were effective.

Precision targeting enables resource efficiency: By identifying specific plant targets (lettuce, chickweed, meadowgrass) and quantifying exact spray deposit amounts (measured in microliters), this technology could help reduce chemical usage by ensuring only necessary plants are treated and only necessary amounts are applied, with the paper achieving an average measurement accuracy of 156.8 μL difference.

Publicly available datasets support responsible AI development: The authors made their curated dataset publicly available, which supports transparency and allows other researchers to validate, improve, or audit the system—a best practice for responsible AI development in agriculture.

Weak supervision enables practical real-world application: The use of "weakly supervised" learning (training with incomplete labels) makes the system more practical for real farms where obtaining perfectly labeled training data would be expensive or impractical, lowering barriers to adoption.