Answers are grounded in this paper and cite their sources.
Traditional approaches to controlling cigar tobacco leaf curing environments rely heavily on the experiential knowledge of master technicians. This study proposes a vision-based multi-target predictive methodology that leverages deep learning to infer environmental conditions, curing stages, and control states from visual information captured during the curing process. We designed and deployed a comprehensive IoT monitoring system with distributed sensors throughout commercial curing facilities, serving as a ground truth annotation system to generate high-quality labeled datasets. Based on 1,003 samples from five curing batches, we developed a ResNet-50-based multi-task learning model that predicts 20 environmental parameters (regression, RMSE: 0.26–0.29), 8 binary control states (accuracy: 67.7%), and 3 curing stage classes (accuracy: 55.0%) from a single image input. We acknowledge limitations including the relatively small dataset size, the absence of temporal modeling, and the lack of baseline comparisons, which constrain the generalizability and practical deployment readiness of current results. The proposed system demonstrates the feasibility of vision-based process monitoring and provides a foundation for intelligent curing assistance, with future work focusing on expanded datasets, temporal modeling, and comparative evaluations. We position this work as an initial exploration of computer vision for in-process curing monitoring, recognizing that the performance characteristics represent a starting point rather than a deployable solution.
No takes yet. Share an insight, caveat, or question.
Lai et al. (2026) studied this question.
Synapse has enriched 10 related papers. Consider them for comparative context: