Research on Encrypted Transmission and Recognition of Garbage Images in Low-Illumination Environments

Zhenwei Lv, Yapeng Diao, Chunnian Zeng, Weiping Wang, Shufan AnElectronics 2026 · DOI

TL;DR

This paper tackles the challenge of accurately identifying garbage in low-light conditions, a crucial task for smart city waste management. They propose a novel framework that combines a specialized low-light image enhancement network with an attention mechanism to improve both image quality and recognition accuracy, outperforming existing methods.

What Problem Does This Paper Address?

Smart city initiatives increasingly rely on computer vision for automated waste management, particularly for garbage sweepers. However, a significant hurdle is the performance degradation of recognition systems in low-illumination environments. Insufficient light leads to blurry, distorted images, making it difficult for algorithms to discern waste types. This not only impacts the efficiency of cleaning operations but also hinders efforts towards better urban environmental management. Traditional image recognition models often assume good lighting, making them unreliable in real-world scenarios where lighting conditions fluctuate due to time of day, weather, or artificial light sources. While some solutions involve adding extra lighting or using high-quality cameras, these can be costly or impractical. This paper aims to bridge this gap by developing a robust garbage recognition system that excels even in challenging low-light conditions.

Methods

The core of the proposed solution is a two-pronged approach:

  1. Multi-Branch Low-Light Enhancement Network (MBLLEN) with Attention:
    • MBLLEN: This network is designed to enhance low-light images. It features a multi-branch architecture that processes image features in parallel, allowing for a more comprehensive capture of visual information.
  2. Convolutional Block Attention Module (CBAM): To address the issue of local over-enhancement and guide the network's focus, CBAM is integrated. CBAM combines channel and spatial attention mechanisms. This allows the model to selectively focus on important features within different channels and spatial regions of the image, effectively highlighting relevant details while suppressing noise and over-enhanced areas. This attention mechanism is strategically placed within the feature extraction module of the MBLLEN.
  3. Deep Learning-Based Garbage Classification:
    • The enhanced images from the MBLLEN are then fed into a deep learning model for garbage classification. The paper mentions a feature extractor (ResNet101 backbone) and a classifier (fully connected layers with Softmax) as part of the overall garbage recognition framework.
  4. Lightweight Encryption Mechanism:
    • For practical deployment, a lightweight encryption mechanism is considered to ensure secure data transmission, although the details of this mechanism are not elaborated upon in the provided text.

Datasets

  • Synthetic Low-Light Data: The MBLLEN was trained using synthetic low-light images generated from the PASCAL Visual Object Classes (PASCALVOC) dataset. These synthetic images were created by applying gamma correction for brightness degradation and adding Poisson-distributed noise to simulate low-light conditions.
  • Self-Collected TrashBig Dataset: For garbage recognition and evaluation, the authors used a self-collected dataset called TrashBig. This dataset contains eight categories of garbage (battery, cardboard, drugs, glass, metal, paper, plastic, and vegetable) and comprises approximately 4000 images for training and 480 images for validation. The dataset is designed to represent real-world scenarios with variations in lighting, backgrounds, and occlusion, and aims to reduce category bias.

Key Insights and Ideas

  • Synergy of Enhancement and Attention: The paper strongly suggests that combining low-light image enhancement with attention mechanisms is more effective than using either technique in isolation for garbage recognition. The attention mechanism acts as a crucial guide, ensuring that the enhancement process focuses on relevant features and avoids detrimental over-enhancement.
  • Targeted Feature Learning: CBAM's ability to focus on specific channels and spatial regions is key. This allows the model to learn more discriminative features, even when the overall image quality is compromised by low light.
  • Addressing Real-World Challenges: The work directly confronts the practical limitations of current vision systems in urban environments, where consistent, ideal lighting is rare.
  • Importance of Domain-Specific Datasets: The creation and use of the TrashBig dataset highlight the need for specialized benchmarks to accurately evaluate models designed for specific applications like waste management.

Key Results

  • Improved Image Quality: The proposed MBLLEN with CBAM demonstrated enhanced image quality, with clearer edges and more discernible details in low-light conditions, as shown in visual examples.
  • Quantitative Performance Gains:
    • In low-light enhancement, the model with the attention mechanism achieved a 3% higher PSNR compared to the model without it.
  • On the TrashBig dataset, the proposed garbage recognition model achieved the highest accuracy and F1-score among the compared methods (Faster R-CNN, Swin-T, and Dark-Waste), with an F1-score of 0.74.
  • The paper also presents detailed AP (Average Precision) scores for various garbage categories when the Illumination Conversion (IC) module (likely referring to the MBLLEN + CBAM) was integrated with different models. For instance, the proposed method with IC achieved 89.5% AP for "battery" and 80.1% AP for "cardboard".

Limitations and Future Work

  • Environmental Complexity: The authors acknowledge that extreme illumination variations, strong glare, and deep shadows can still limit the recoverability of visual information and affect recognition accuracy.
  • Object Occlusion and Variability: Real-world scenarios often involve object occlusion, overlapping waste items, and significant intra-class variability (shape, color, size), which pose ongoing challenges.
  • Dataset Limitations: Class imbalance and annotation noise in datasets can restrict model generalization.
  • Future Directions: The paper suggests that future improvements should focus on developing more robust enhancement strategies and better methods for handling real-world data uncertainties. This includes exploring more advanced enhancement techniques and more sophisticated ways to manage complex environmental factors and dataset-related issues.

Summary updated

Abstract

Low-illumination conditions significantly degrade the performance of vision-based garbage recognition systems in practical smart city applications. To address this issue, this paper presents a garbage recognition framework that combines low-light image enhancement with attention-guided feature learning. A multi-branch low-light enhancement network (MBLLEN) is employed as the enhancement backbone, and a Convolutional Block Attention Module (CBAM) is integrated to alleviate local over-enhancement and guide feature responses under uneven illumination. The enhanced images are then used as inputs for a deep learning-based garbage classification model. In addition, a lightweight encryption mechanism is considered at the system level to support secure data transmission in practical deployment scenarios. Experiments conducted on a self-collected low-light garbage dataset show that the proposed framework achieves improved image quality and recognition performance compared with baseline approaches. These results suggest that integrating low-light enhancement with attention-guided feature learning can be beneficial for garbage recognition tasks under challenging illumination conditions.
BibTeX
@inproceedings{lv2026research,
  author = {Zhenwei Lv and Yapeng Diao and Chunnian Zeng and Weiping Wang and Shufan An},
  title = {{Research on Encrypted Transmission and Recognition of Garbage Images in Low-Illumination Environments}},
  booktitle = {Electronics},
  year = {2026},
  doi = {10.3390/electronics15020302},
}
Figure