声明
摘要
Abstract
1 Introduction
1.1 Research purpose and significance
1.2 Domestic and international research status
1.2.1 Object detection and recognition research status
1.2.2 Cattle detection and recognition research status
1.2.3 Cattle detection and recognition applications research status
1.3 Research content and chapters arrangement
1.3.1 Research content
1.3.2 Chapters arrangement
2 Related technologies
2.1 TensorFlow
2.1.1 About TensorFlow
2.1.2 TensorFlow GPU and CPU
2.2 Single Shot MultiBox Detector (SSD) based on MobileNet
2.2.1 MobileNet feature extraction
2.2.2 Single Shot MultiBox Detector structure
2.2.3 Network training
2.3 TensorFlow on Android device
2.3.1 TensorFlow mobile
2.3.2 TensorFlow lite
3 Development environment and configuration utilities
3.1 Work environment
3.2 TensorFlow object detection API
3.3 Python
3.4 CUDA and cuDNN
3.5 Deploying in Android
3.5.1 Creating an Android application
3.5.2 Benchmarking TensorFlow on Android devices
4 Neural network training process
4.1 Hardware
4.2 Dataset creation
4.3 Neural network model configuration
4.3.1 Generate training data
4.3.2 Create label map
4.3.3 Configuring a training pipeline
4.4 Training model and testing process
4.4.1 Running the training
4.4.2 Monitor training progress using TensorBoard
4.4.3 Export inference graph
4.4.4 Test the model
4.5 Running a neural network on Android
5 Experimental results and analysis
5.1 Training analysis
5.2 Mean Average Precision(mAP)performance
5.3 Detection and recognition testing and results
5.3.1 Testing and results on the PC
5.3.2 Testing and results on Android device
5.4 Discussion
6 Conclusion and Future work
6.2 Future work
Acknowledgment
References
东北农业大学;