EcoDetect-YOLO: A Lightweight, High-Generalization Methodology for Real-Time Detection of Domestic Waste Exposure in Intricate Environmental Landscapes

Abstract

In response to the challenges of accurate identification and localization of garbage in intricate urban street environments, this paper proposes EcoDetect-YOLO, a garbage exposure detection algorithm based on the YOLOv5s framework, utilizing an intricate environment waste exposure detection dataset constructed in this study. Initially, a convolutional block attention module (CBAM) is integrated between the second level of the feature pyramid etwork (P2) and the third level of the feature pyramid network (P3) layers to optimize the extraction of relevant garbage features while mitigating background noise. Subsequently, a P2 small-target detection head enhances the model's efficacy in identifying small garbage targets. Lastly, a bidirectional feature pyramid network (BiFPN) is introduced to strengthen the model's capability for deep feature fusion. Experimental results demonstrate EcoDetect-YOLO's adaptability to urban environments and its superior small-target detection capabilities, effectively recognizing nine types of garbage, such as paper and plastic trash. Compared to the baseline YOLOv5s model, EcoDetect-YOLO achieved a 4.7% increase in mAP_0.5, reaching 58.1%, with a compact model size of 15.7 MB and an FPS of 39.36. Notably, even in the presence of strong noise, the model maintained a mAP_0.5 exceeding 50%, underscoring its robustness. In summary, EcoDetect-YOLO, as proposed in this paper, boasts high precision, efficiency, and compactness, rendering it suitable for deployment on mobile devices for real-time detection and management of urban garbage exposure, thereby advancing urban automation governance and digital economic development.

Keywords: BiFPN; CBAM; P2; domestic waste; intricate environmental landscapes; lightweight.

Figure 1

The detection architecture of YOLOv5 features input, backbone, neck, and head.

Figure 2

The overall network architecture of YOLOv5s.

Figure 3

Example of the Multi-Target Life Garbage Exposure Image Dataset.

Figure 4

Dataset distribution: the blue bar chart represents the distribution of the number of images for each type of garbage; the red bar chart represents the distribution of the number of each type of garbage.

Figure 5

Results of adaptive scaling applied to original images.

Figure 6

Results of HSV color-space augmentation and random perspective augmentation. (a) Original image, (b) image after HSV color-space augmentation, and (c–f) images after random perspective augmentation with adjustments in degrees, scale, shear, and translate, respectively.

Figure 7

Results of mosaic augmentation and MixUp augmentation. (a) Image after mosaic augmentation and (b) image after MixUp augmentation.

Figure 8

Structure diagram of the CBAM attention mechanism.

Figure 9

CBAM channel attention module.

Figure 10

CBAM spatial attention module.

Figure 11

Network architecture diagram of the model. (a) YOLOv5s model. (b) YOLOv5s+P2 model.

Figure 11

Network architecture diagram of the model. (a) YOLOv5s model. (b) YOLOv5s+P2 model.

Figure 12

Feature maps of the YOLOv5s and YOLOv5s+P2 models. (a) Original image, (b) 2nd layer feature map, (c) 17th layer feature map of the YOLOv5s model, and (d) 21st layer feature map of the YOLOv5s+P2 model. The red boxes are used to mark the positions of the three samples.

Figure 12

Feature maps of the YOLOv5s and YOLOv5s+P2 models. (a) Original image, (b) 2nd layer feature map, (c) 17th layer feature map of the YOLOv5s model, and (d) 21st layer feature map of the YOLOv5s+P2 model. The red boxes are used to mark the positions of the three samples.

Figure 13

Testing results of the YOLOv5s+P2 and YOLOv5s models. (a) YOLOv5s+P2 model testing result. (b) YOLOv5s model testing result.

Figure 14

BiFPN network architecture diagram. (a) FPN and PANet structures in the YOLOv5s model after addition the P2 small-object detection layer. (b) BiFPN structure in the YOLOv5s model after addition of the P2 small-object detection layer.

Figure 15

EcoDetect-YOLO overall network architecture.

Figure 16

Confusion matrix of EcoDetect-YOLO and Yolov5s in the training set. (a) Confusion matrix of EcoDetect-YOLO. (b) Confusion matrix of Yolov5s.

Figure 17

Detection results of the EcoDetect-YOLO and YOLOv5s models. (a) The test results of the proposed EcoDetect-YOLO model. (b) The test results of the baseline YOLOv5s model. Blue boxes added to the figures represent the missed detection of target samples by the model, while green boxes indicate incorrectly detected target samples.

Figure 17

Detection results of the EcoDetect-YOLO and YOLOv5s models. (a) The test results of the proposed EcoDetect-YOLO model. (b) The test results of the baseline YOLOv5s model. Blue boxes added to the figures represent the missed detection of target samples by the model, while green boxes indicate incorrectly detected target samples.

Figure 17

Detection results of the EcoDetect-YOLO and YOLOv5s models. (a) The test results of the proposed EcoDetect-YOLO model. (b) The test results of the baseline YOLOv5s model. Blue boxes added to the figures represent the missed detection of target samples by the model, while green boxes indicate incorrectly detected target samples.

Figure 18

Performance comparison of the different detection algorithms.

Figure 19

Robustness test results. (a) The original image. (b) The image with increased brightness, (c) The image with decreased brightness. (d) The image with added Gaussian noise.