Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

An improved EfficientNetV2 for garbage classification

Wenxuan Qiu, Chengxin Xie, Jingui Huang

TL;DR

This work tackles waste classification under resource constraints by integrating a Channel-Efficient (CE) attention mechanism and a depthwise-separable multi-scale feature module (DP-SAFM) into EfficientNetV2, complemented by comprehensive data augmentation. The CE attention enhances fine-grained channel discrimination, while DP-SAFM strengthens spatial feature modeling with lower computational cost. Experimental results on Huawei Cloud and TrashNet demonstrate a 3.2% accuracy gain over baselines, with 95.4% on Huawei Cloud and 96.5% on TrashNet, confirming a favorable balance between accuracy and efficiency suitable for edge deployment. Overall, the approach advances robust, real-time waste classification in cluttered and diverse environments, addressing data scarcity and generalization challenges in practical applications.

Abstract

This paper presents an enhanced waste classification framework based on EfficientNetV2 to address challenges in data acquisition cost, generalization, and real-time performance. We propose a Channel-Efficient Attention (CE-Attention) module that mitigates feature loss during global pooling without introducing dimensional scaling, effectively enhancing critical feature extraction. Additionally, a lightweight multi-scale spatial feature extraction module (SAFM) is developed by integrating depthwise separable convolutions, significantly reducing model complexity. Comprehensive data augmentation strategies are further employed to improve generalization. Experiments on the Huawei Cloud waste classification dataset demonstrate that our method achieves a classification accuracy of 95.4\%, surpassing the baseline by 3.2\% and outperforming mainstream models. The results validate the effectiveness of our approach in balancing accuracy and efficiency for practical waste classification scenarios.

An improved EfficientNetV2 for garbage classification

TL;DR

This work tackles waste classification under resource constraints by integrating a Channel-Efficient (CE) attention mechanism and a depthwise-separable multi-scale feature module (DP-SAFM) into EfficientNetV2, complemented by comprehensive data augmentation. The CE attention enhances fine-grained channel discrimination, while DP-SAFM strengthens spatial feature modeling with lower computational cost. Experimental results on Huawei Cloud and TrashNet demonstrate a 3.2% accuracy gain over baselines, with 95.4% on Huawei Cloud and 96.5% on TrashNet, confirming a favorable balance between accuracy and efficiency suitable for edge deployment. Overall, the approach advances robust, real-time waste classification in cluttered and diverse environments, addressing data scarcity and generalization challenges in practical applications.

Abstract

This paper presents an enhanced waste classification framework based on EfficientNetV2 to address challenges in data acquisition cost, generalization, and real-time performance. We propose a Channel-Efficient Attention (CE-Attention) module that mitigates feature loss during global pooling without introducing dimensional scaling, effectively enhancing critical feature extraction. Additionally, a lightweight multi-scale spatial feature extraction module (SAFM) is developed by integrating depthwise separable convolutions, significantly reducing model complexity. Comprehensive data augmentation strategies are further employed to improve generalization. Experiments on the Huawei Cloud waste classification dataset demonstrate that our method achieves a classification accuracy of 95.4\%, surpassing the baseline by 3.2\% and outperforming mainstream models. The results validate the effectiveness of our approach in balancing accuracy and efficiency for practical waste classification scenarios.

Paper Structure

This paper contains 16 sections, 9 equations, 6 figures, 4 tables.

Table of Contents

  1. Introduction
  2. Related Work
  3. EfficientNetV2 and Its Limitations
  4. Spatially-Adaptive Feature Modulation (SAFM)
  5. Proposed Method
  6. Model
  7. Channel Efficient(CE) attention module
  8. Improved SAFM with Depthwise Separable Convolutions (DP-SAFM)
  9. Data augmentation
  10. Experiment and Analysis
  11. Datasets
  12. Evaluation Metrics
  13. Implementation Details
  14. Comparison with State-of-the-art Methods
  15. Ablation Studies and Analysis
  16. ...and 1 more sections

Figures (6)

  • Figure 1: Structure of the MBConv and Fused-MBConv module in EfficientNetV2
  • Figure 2: Structure of the CE-EfficientNetV2 Model
  • Figure 3: CE attention module
  • Figure 4: Data Augmentation Strategies
  • Figure 5: Line Charts of Accuracy and Loss Rates for Different Models on the Huawei Cloud Dataset
  • ...and 1 more figures