CNN-Bi-LSTM: A Complex Environment-Oriented Cattle Behavior Classification Network Based on the Fusion of CNN and...

Area of Science:

• Agricultural Technology
• Computer Vision
• Machine Learning

Background:

• Smart cattle farming requires accurate behavior classification for improved animal welfare and management.
• Existing methods struggle with complex farm environments, varying light, and occlusions.

Purpose of the Study:

• To develop and validate a novel deep learning network for precise cattle behavior classification in intricate agricultural settings.
• To enhance the accuracy and generalizability of cattle behavior recognition beyond traditional single-sensor data analysis.

Main Methods:

• A novel network combining VGG16-based CNN for spatial feature extraction and Bi-LSTM for temporal semantic analysis was developed.
• Data was collected in an authentic farm setting, defining eight fundamental cattle behaviors.
• Ablation experiments, generalization assessments, and comparative analyses with MASK-RCNN, CNN-LSTM, and EfficientNet-LSTM were conducted.

Main Results:

• The proposed CNN-Bi-LSTM model achieved 94.3% accuracy, 94.2% precision, and 93.4% recall.
• The model demonstrated superior performance compared to MASK-RCNN, CNN-LSTM, and EfficientNet-LSTM.
• Effective generalization across diverse subjects and viewing perspectives was confirmed.

Conclusions:

• The CNN-Bi-LSTM fusion effectively extracts multimodal features for robust cattle behavior classification in challenging environments.
• This approach significantly improves precision and generalizability, addressing limitations of conventional methods.
• The technology offers substantial practical, economic, and societal benefits for the agricultural sector.