Image-based honeybee colony conditions detection using a hybrid CNN–ANN framework

Automated assessment of honeybee colony conditions from imagery remains challenging due to class imbalance and subtle inter-class visual differences. This work proposes a hybrid deep learning framework that combines dual-branch Convolutional Neural Networks (CNN) with a Multi-Layer Feedback Artificial Neural Network (MLFB-ANN) classifier to enhance feature representation through a single-step feedback mechanism. Experiments conducted on the BeeImage dataset show that the proposed model outperforms a baseline CNN+Softmax architecture across standard metrics. Over 10 independent runs, the method achieves higher accuracy and Macro-F1 with reduced variance, indicating improved stability. Statistical validation using the Wilcoxon signed-rank test (p = 0.00195) and McNemar’s test (p = \(4.23 \times 10^{-66}\) $4.23\times {10}^{-66}$ ) confirms that the observed improvements are significant. These results demonstrate the effectiveness of feedback-enhanced classification within the evaluated dataset. However, as the data originate from curated or semi-controlled conditions, further evaluation on diverse field-acquired imagery is required to assess generalisation under real-world scenarios.