Classifying bird species is essential for ecological study and biodiversity protection, currently, conventional approaches are frequently laborious and susceptible to mistakes. Convolutional Neural Networks (CNNs) provide a more reliable option for feature extraction and classification. By combining the top three independently superior CNN architectures for recognizing bird species from the DenseNet and ResNet families into a fuzzy-based ensemble learning framework, this study helps increase classification accuracy, especially for occluded bird objects. The model improves generalization by using 11,352 images collected from the Caltech-UCSD Birds-200-2011 and Birds525 Species-Image Classification datasets, as well as sophisticated augmentation approaches. Our ensemble method adaptively allocates model weights based on feature contributions found using fuzzy logic, in contrast to existing methods that have trouble with obstructed images. Since, every CNN model candidate in the suggested fuzzy-based ensemble learning showed excellent classification performance, the proposed fuzzy-based ensemble approach achieving 98.73% accuracy, a 98.75% F1-score for standard images, and 95.78% accuracy, and a 95.1% F1-score for occluded images, the results indicating performance improvements of 2% for standard and 9% for occluded bird images over the methods used in existing research work. Furthermore, as compared to the individual CNN candidates in the proposed fuzzy-based ensemble, this indicates a 2-5% performance improvement for standard bird images and a 4-7% performance improvement for occluded bird images. Additionally, the reliability and significance of the observed performance increases are verified by statistical validation of the results using p-value and F-statistic testing and 95% Confidence Intervals.