Improved CNN for Classifying Floating Objects in Water
Author Information
Author(s): Yang Jikai, Li Zihan, Gu Ziyan
Primary Institution: Huazhong University of Science and Technology
Hypothesis
Can an improved VGG-16 model effectively classify floating objects on water surfaces using deep learning?
Conclusion
The improved VGG16-15 model achieved a recognition accuracy of 93.86%, demonstrating significant improvements over traditional models.
Supporting Evidence
- The model achieved a recognition accuracy of 93.86%, a 10.09% improvement over the traditional VGG-16 model.
- Data augmentation techniques significantly enhanced the model's generalization ability.
- The few-shot test showed the fine-tuned model's superior performance with limited data.
Takeaway
This study created a smart system that helps unmanned boats recognize and sort trash in water using pictures, making it easier to keep our waters clean.
Methodology
The study used a convolutional neural network (CNN) with a modified VGG-16 architecture to classify 15 types of floating debris from a dataset of 5707 images.
Potential Biases
Potential bias due to imbalanced sample sizes across different categories.
Limitations
The model may struggle with certain categories due to limited sample diversity and may not perform well under varying environmental conditions.
Statistical Information
P-Value
0.0001
Statistical Significance
p<0.05
Digital Object Identifier (DOI)
Want to read the original?
Access the complete publication on the publisher's website