Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

YOLO-SAT: A Data-based and Model-based Enhanced YOLOv12 Model for Desert Waste Detection and Classification

Abdulmumin Sa'ad, Sulaimon Oyeniyi Adebayo

TL;DR

This work tackles automated desert waste detection by enhancing a lightweight YOLOv12n with data-centric augmentations and model-centric pruning plus Self-Adversarial Training (SAT). By training on the DroneTrashNet dataset and applying Mosaic/CutMix augmentations, negative samples, and a careful pruning strategy, the authors achieve a high-precision, low-latency detector suitable for embedded drones. SAT mitigates accuracy loss from pruning and improves robustness to desert conditions, delivering mAP@0.50:0.95 of 0.7783 with a 16.3 ms latency and 2.17M parameters. The results demonstrate a favorable accuracy–efficiency balance compared to other lightweight YOLO variants, enabling robust, real-time desert waste detection in resource-constrained aerial platforms.

Abstract

The global waste crisis is escalating, with solid waste generation expected to increase tremendously in the coming years. Traditional waste collection methods, particularly in remote or harsh environments like deserts, are labor-intensive, inefficient, and often hazardous. Recent advances in computer vision and deep learning have opened the door to automated waste detection systems, yet most research focuses on urban environments and recyclable materials, overlooking organic and hazardous waste and underexplored terrains such as deserts. In this work, we propose YOLO-SAT, an enhanced real-time object detection framework based on a pruned, lightweight version of YOLOv12 integrated with Self-Adversarial Training (SAT) and specialized data augmentation strategies. Using the DroneTrashNet dataset, we demonstrate significant improvements in precision, recall, and mean average precision (mAP), while achieving low latency and compact model size suitable for deployment on resource-constrained aerial drones. Benchmarking YOLO-SAT against state-of-the-art lightweight YOLO variants further highlights its optimal balance of accuracy and efficiency. Our results validate the effectiveness of combining data-centric and model-centric enhancements for robust, real-time waste detection in desert environments.

YOLO-SAT: A Data-based and Model-based Enhanced YOLOv12 Model for Desert Waste Detection and Classification

TL;DR

This work tackles automated desert waste detection by enhancing a lightweight YOLOv12n with data-centric augmentations and model-centric pruning plus Self-Adversarial Training (SAT). By training on the DroneTrashNet dataset and applying Mosaic/CutMix augmentations, negative samples, and a careful pruning strategy, the authors achieve a high-precision, low-latency detector suitable for embedded drones. SAT mitigates accuracy loss from pruning and improves robustness to desert conditions, delivering mAP@0.50:0.95 of 0.7783 with a 16.3 ms latency and 2.17M parameters. The results demonstrate a favorable accuracy–efficiency balance compared to other lightweight YOLO variants, enabling robust, real-time desert waste detection in resource-constrained aerial platforms.

Abstract

The global waste crisis is escalating, with solid waste generation expected to increase tremendously in the coming years. Traditional waste collection methods, particularly in remote or harsh environments like deserts, are labor-intensive, inefficient, and often hazardous. Recent advances in computer vision and deep learning have opened the door to automated waste detection systems, yet most research focuses on urban environments and recyclable materials, overlooking organic and hazardous waste and underexplored terrains such as deserts. In this work, we propose YOLO-SAT, an enhanced real-time object detection framework based on a pruned, lightweight version of YOLOv12 integrated with Self-Adversarial Training (SAT) and specialized data augmentation strategies. Using the DroneTrashNet dataset, we demonstrate significant improvements in precision, recall, and mean average precision (mAP), while achieving low latency and compact model size suitable for deployment on resource-constrained aerial drones. Benchmarking YOLO-SAT against state-of-the-art lightweight YOLO variants further highlights its optimal balance of accuracy and efficiency. Our results validate the effectiveness of combining data-centric and model-centric enhancements for robust, real-time waste detection in desert environments.

Paper Structure

This paper contains 12 sections, 6 figures, 3 tables.

Table of Contents

  1. Introduction
  2. Related Work
  3. Methodology
  4. Experimental Setup
  5. Model Architecture
  6. Dataset Generation
  7. Impact of Self-Adversarial Training (SAT)
  8. Hardware and Software Setup
  9. Results and Discussion
  10. Evaluation and Results
  11. Benchmarking
  12. Conclusion

Figures (6)

  • Figure 1: Categories of Waste.
  • Figure 2: Engineering Application Scenario of Desert Waste Detection System.
  • Figure 3: Proposed Data-Based and Model-Based Optimization Framework of YOLO-SAT.
  • Figure 4: Redefined YOLOv12n model architecture with SAT.
  • Figure 5: Single-class detection across: (a) off-the-shelf YOLOv12n, (b) raw-trained, (c) noisy-trained, (d) augmented-trained.
  • ...and 1 more figures