Image Representation Method Based on Relative Layer Entropy for Insulator Recognition.

Area of Science:

• Computer Vision
• Machine Learning
• Deep Learning

Background:

• Deep convolutional neural networks (DCNNs) are computationally intensive.
• Selecting optimal feature representation layers in DCNNs remains a challenge.
• Limited training data and object appearance types hinder recognition tasks.

Purpose of the Study:

• To propose a novel image representation method based on relative layer entropy (IRM_RLE).
• To identify the most suitable convolution layer for image recognition.
• To enhance recognition performance with improved feature selection.

Main Methods:

• Utilized an ImageNet pretrained DCNN to extract deep convolutional activations.
• Introduced layer entropy and relative layer entropy to select optimal feature layers.
• Employed VLAD coding for feature map vectorization and quantification.

Main Results:

• The proposed IRM_RLE method achieves competitive performance across various datasets.
• IRM_RLE outperforms state-of-the-art methods on indoor scenes and actions datasets.
• The method effectively identifies crucial feature layers for image representation.

Conclusions:

• IRM_RLE offers an effective strategy for selecting optimal feature layers in DCNNs.
• The approach addresses computational complexity and improves recognition accuracy.
• This method provides a valuable tool for feature representation in computer vision.