Fast DCNN based on FWT, intelligent dropout and layer skipping for image retrieval

Area of Science:

• Computer Vision
• Machine Learning
• Deep Learning

Background:

• Deep Convolutional Neural Networks (DCNNs) are effective for image classification and retrieval but require significant computational resources for training.
• Optimization of DCNN training remains a complex challenge, impacting efficiency and applicability.

Purpose of the Study:

• To propose a computationally efficient Deep Convolutional Neural Network (DCNN) architecture for accelerated image retrieval and classification.
• To address the time and storage demands associated with training conventional DCNNs.

Main Methods:

• The proposed method integrates Fast Wavelet Transform (FWT) for rapid feature computation.
• An intelligent dropout technique is employed, selecting units based on efficiency rather than random chance.
• Layer skipping allows for direct classification by utilizing earlier layers and bypassing subsequent ones.

Main Results:

• The developed DCNN model demonstrated improved image retrieval accuracy.
• Significant reduction in image searching time was achieved.
• Experiments on CIFAR-10 and MNIST datasets yielded promising results, validating the approach.

Conclusions:

• The combination of FWT, intelligent dropout, and layer skipping offers a viable solution for faster and more accurate DCNN-based image retrieval.
• This approach effectively mitigates the computational burden of DCNN training.
• The method shows potential for practical applications requiring efficient image analysis.