عنوان المشروع: PALMY: An AI-Powered UAV System for Automated Date Palm Disease Detection Using Fine-Tuned Vision-Language Models
الوصف:
Manual date palm inspection is costly, slow, and physically dangerous. This paper presents PALMY, a UAV-based system leveraging a fine-tuned vision-language model (Qwen3-VL-2B-Instruct) for automated aerial disease detection. The key innovation is a self-correcting annotation pipeline that enables low-cost dataset generation, coupled with a VLM architecture that replaces black-box CNN labels with structured natural-language diagnostic reports. Using LoRA (r = 8) on a 4-bit quantized base model, PALMY was fine-tuned for 30 steps (~5 epochs) on free-tier cloud hardware. Training loss converged to 0.761 from an initial 4.64, an 83.6% reduction, across 11 target classes spanning diseases, pests, nutrient deficiencies, and healthy status. The system outputs Excel reports with severity, urgency, and treatment recommendations. Validation with government officials and farmers confirms PALMY's practical agricultural impact. As a proof-of-concept, PALMY Phase 1 demonstrates pipeline feasibility rather than production readiness, with full-scale validation scoped to future work.
تاريخ رفع الملفات: 2026-07-05 08:28:36