WeedExpert-R1: Incentivizing Botanical Reasoning in MLLMs with Reinforcement Learning for Precision Weed Grounding
Computer Vision and Pattern Recognition
Summary
The authors developed WeedExpert-R1, a new AI model that can identify and locate different weed species in farming images, even those it hasn’t seen before. Unlike traditional models that only recognize fixed categories, this model uses language and botanical traits to better understand and explain its decisions. They created a special training process combining plant knowledge with AI reasoning to improve accuracy. Tests showed that WeedExpert-R1 outperforms other advanced models and works well across various plants and farming areas without needing retraining.
Authors
Zonglin Yang, Wei-Zhen Liang, Nevin Lawrence, Xin Qiao, Benjamin Riggan, Chi-En Chiang, Fuchen Li
Abstract
Precision weed control requires species-level identification and instance-level localization. However, conventional object detectors use a closed vocabulary, limiting their deployment across regions, and cannot explain their predictions in complex agricultural scenes. Multimodal large language models (MLLMs) offer visual grounding and reasoning capabilities, but insufficient botanical knowledge can cause hallucinations in fine-grained weed identification. This study introduces WeedExpert-R1, a multimodal model that learns visually grounded botanical reasoning through verifiable rewards. A domain-specific Chain-of-Thought synthesis pipeline combines a human-curated botanical trait dictionary with an Auditor-Synthesizer LLM workflow to generate reasoning data for supervised fine-tuning. Group Relative Policy Optimization is then applied with rewards for format, accuracy, instance count, and response length. Across 37 weed species from six datasets, WeedExpert-R1-4B achieved 75.82 percent exact-set precision at an IoU threshold of 0.5, 89.30 percent precision, and 87.81 percent recall. It outperformed proprietary models, including GPT-5.4 and Gemini-3.1-Pro, and larger open-source models, including Qwen3-VL-30B-Instruct and Gemma-4-31B-it. Results on unseen species further demonstrate its open-vocabulary capability and potential for deployment across diverse regions and crops without retraining.