Consecutive multiscale feature learning-based image classification model.

Area of Science:

• Computer Vision
• Deep Learning
• Image Classification

Background:

• Multiscale feature extraction is vital in computer vision.
• Current parallel CNN approaches struggle with efficiency and generalization on small-scale images.
• Lightweight networks often underfit small datasets.

Purpose of the Study:

• To propose a novel image classification system addressing limitations of current methods.
• To introduce the consecutive multiscale feature-learning network (CMSFL-Net).
• To improve efficiency, speed, and accuracy in image classification.

Main Methods:

• Developed a novel CNN architecture, CMSFL-Net.
• Employed a consecutive feature-learning approach with varied receptive fields.
• Utilized elaborate data preprocessing techniques.

Main Results:

• CMSFL-Net achieved accuracy comparable to state-of-the-art efficient networks.
• The proposed system demonstrated superior efficiency and speed.
• CMSFL-Net excelled in the accuracy-efficiency trade-off across diverse datasets.

Conclusions:

• CMSFL-Net offers a promising solution for efficient and accurate multiscale feature extraction.
• The consecutive learning approach improves performance on small-scale and limited data.
• CMSFL-Net provides a competitive accuracy-efficiency balance for image classification.