Atmospheric Noise-Resilient Image Classification in a Real-World Scenario: Using Hybrid CNN and Pin-GTSVM

TL;DR

This work targets parking-slot classification under atmospheric noise, introducing a hybrid model that couples pre-trained CNN feature extractors (e.g., ResNet-50, AlexNet, GoogLeNet) with a Pin-GTSVM classifier to achieve robust, dehaze-free occupancy detection. By leveraging Pinball loss within a twin-SVM framework and kernel mappings, the approach enhances resilience to haze while maintaining real-world deployability with minimal camera infrastructure. The method is evaluated on PKLot, CNRPark, and a hazy-custom dataset, showing superior accuracy (up to 98.20%) and strong generalization across conditions, while eliminating the need for separate dehazing steps. The results suggest practical impact for smart parking systems, enabling reliable operation in adverse weather and enabling efficient scaling with fewer sensors and reduced preprocessing overhead.

Abstract

Parking space occupation detection using deep learning frameworks has seen significant advancements over the past few years. While these approaches effectively detect partial obstructions and adapt to varying lighting conditions, their performance significantly diminishes when haze is present. This paper proposes a novel hybrid model with a pre-trained feature extractor and a Pinball Generalized Twin Support Vector Machine (Pin-GTSVM) classifier, which removes the need for a dehazing system from the current State-of-The-Art hazy parking slot classification systems and is also insensitive to any atmospheric noise. The proposed system can seamlessly integrate with conventional smart parking infrastructures, leveraging a minimal number of cameras to monitor and manage hundreds of parking spaces efficiently. Its effectiveness has been evaluated against established parking space detection methods using the CNRPark Patches, PKLot, and a custom dataset specific to hazy parking scenarios. Furthermore, empirical results indicate a significant improvement in accuracy on a hazy parking system, thus emphasizing efficient atmospheric noise handling.

Figures (20)

• Figure 1: Pinball Loss Functionpin78
• Figure 2: Res-Net50 Architecture
• Figure 3: Overview of the AlexNet Architecture AlexNetimg
• Figure 4: GoogleNet Architecture
• Figure 5: (a) Parking spot containing 28 delimited parking spaces, (b) occupied parking slot, and (c) empty parking slot.