Entropy-driven adaptive spatial modeling of multi-stress responses in plant systems under environmental variability

Environmental stresses, including drought, salinity, heat, and nutrient deficiency, can alter plant physiology, morphology, and growth. Reliable analysis of plant responses to co-occurring environmental stresses remains challenging because stress-associated visual patterns may overlap and vary under changing environmental conditions. This study proposes an Entropy-Driven Adaptive Spatial Modeling (ASM) framework for analyzing stress-associated patterns in RGB grapevine leaf images. First, Adaptive Gabor Filtering is applied to enhance orientation-sensitive structural features while reducing background noise and illumination-related variations. Next, Histogram of Oriented Gradients and multi-scale feature extraction are used to represent edge, texture, and morphological characteristics. A Swin Transformer-assisted Capsule Network is then employed to learn hierarchical spatial representations, while a Graph Attention Network models relationships among localized image-derived regions. Particle Swarm Optimization is used to tune selected model hyperparameters. Finally, entropy analysis is applied to quantify uncertainty and complexity in the predicted stress-response distribution. Experimental results indicate that the proposed framework achieves higher classification performance than the selected baseline models. The framework provides an image-based approach for interpreting stress-associated visual patterns under environmental variability.