Accurate plant leaf identification is critical for biodiversity monitoring, agricultural productivity, and disease diagnosis. A key limitation of the Densenet121 architecture is the lack of transparency within its intermediate layers during classification, making it difficult to analyze or interpret the training process. Conversely, Random Vector Functional Link (RVFL) networks provide efficient learning but are highly sensitive to noise. To address these limitations, we propose a novel hybrid framework that integrates DenseNet121 for deep feature extraction with an Intuitionistic Fuzzy Random Vector Functional Link (IFRVFL) classifier for robust decision-making. DenseNet121 captures rich hierarchical representations of leaf structures, while IFRVFL leverages fuzzy membership and non-membership values to mitigate the impact of noisy and uncertain samples. Extensive experiments were conducted on multiple benchmark plant leaf datasets. The proposed DenseNet121–IFRVFL consistently outperformed baseline models, including DenseNet121-SVM, DenseNet121-TSVM, DenseNet121-ELM, DenseNet121-KRR, and DenseNet121-RVFL. Our model achieved superior performance across Accuracy (95.20%), AUC (0.964), F1-score (0.969), and G-Mean (0.924), demonstrating both robustness to imbalanced data and resistance to noise. The results confirm that combining deep feature extraction with fuzzy noise-resilient classification offers a powerful solution for practical plant identification tasks.