<![CDATA[Mushroom species identification based on digital images remains challenging due to high visual similarity among species and the limitations of manual identification. This study proposes a Hybrid Fusion Deep Learning classification system that integrates two Convolutional Neural Network (CNN) architectures, DenseNet201 and MobileNetV3Large, as feature extractors through feature-level fusion. The novelty of this research lies in the complementary integration of two CNN architectures not previously combined for mushroom classification, further coupled with Random Forest as the final classifier to improve stability and generalization. The dataset comprises 2,573 images of five mushroom classes from Kaggle, split at 70:15:15 (training: 1,801; validation: 386; testing: 386 images). Preprocessing includes resizing to 224×224 pixels, pixel normalization, and data augmentation. Evaluation was conducted using stratified 5-fold cross-validation, with accuracy, precision, recall, F1-score, and AUC metrics. The proposed model achieves a validation accuracy of 95.53%, with micro-average AUC = 0.9961 and macro-average AUC = 0.9951. Compared to single-model baselines (DenseNet201: 91.45%; MobileNetV3Large: 89.73%), the proposed method demonstrates significant improvement. These findings confirm that the hybrid fusion approach effectively enhances mushroom image classification and has strong potential for computer vision-based biological identification in food safety and biodiversity conservation.]]>
