Exploration of MPSO-Two-Stage Classification Optimization Model for Scene Images with Low Quality and Complex...

Area of Science:

• Computer Vision
• Machine Learning
• Image Processing

Background:

• Current complex scene classification methods primarily focus on high-definition images, neglecting low-quality and low-resolution datasets.
• Existing research on noisy images or specific sets does not address actual low-resolution scene images, limiting practical applications.

Purpose of the Study:

• To propose a practical, high-precision classification model for low-quality scene images.
• To develop a two-stage classification optimization algorithm based on MPSO (Multi-Particle Swarm Optimization).

Main Methods:

• A two-stage classification optimization algorithm model based on MPSO was proposed.
• Comparative experiments were conducted using internationally recognized scene datasets and self-built low-quality news scene frames.
• Adaptive tests were performed on scene sets affected by lighting and weather conditions.

Main Results:

• The proposed MPSO model achieved higher accuracy than 21 existing methods, notably improving performance on a 15-scene dataset by 1.54% compared to ResNet-ELM.
• The pre-reconstruction stage of the model showed a significant improvement rate, particularly for outdoor scenes, when compared to six existing preprocessing methods.
• The model demonstrated suitability for weather-affected scene classification, yielding an average accuracy improvement of 1.42% in adaptive tests.

Conclusions:

• The proposed two-stage MPSO model offers a practical and effective solution for high-precision classification of low-quality scene images.
• The model's pre-reconstruction stage is crucial for enhancing performance, especially in challenging outdoor and variable weather conditions.
• This research advances the field by addressing the overlooked area of low-quality scene image classification.