Optics and Precision Engineering, Volume. 33, Issue 12, 1984(2025)
Improved lightweight garbage detection method for YOLOv8n in complex environments
Figures & Tables(20)
Fig. 11. Comparison of model accuracy and positioning loss curve before and after improvemen
Table 1. MobileNet V3-Small network structure
View tableView in ArticleTable 1. MobileNet V3-Small network structure
操作 输入通道 输入大小 激活函数 SE Conv2d， 3×3 3 224×224 HS - Bneck， 3×3 16 112×112 RE √ Bneck， 3×3 16 56×56 RE - Bneck， 3×3 24 28×28 RE √ Bneck， 5×5 24 28×28 HS √ Bneck， 5×5 40 14×14 HS √ Bneck， 5×5 40 14×14 HS √ Bneck， 5×5 40 14×14 HS √ Bneck， 5×5 48 14×14 HS √ Bneck， 5×5 48 14×14 HS √ Bneck， 5×5 96 7×7 HS √ Bneck， 5×5 96 7×7 HS √ Conv2d， 1×1 96 7×7 HS √ Pool， 7×7 576 7×7 - - Conv2d， 1×1， NBN 576 1×1 HS - Conv2d， 1×1， NBN 1 024 1×1 - -
Table 2. Distribution of images across different scenes
View tableView in ArticleTable 2. Distribution of images across different scenes
背景环境 室内环境 室外环境 草地 沙滩 道路 图像数量/张 823 749 159 811
Table 3. Experimental results of different backbone networks
View tableView in ArticleTable 3. Experimental results of different backbone networks
Models P/% R/% Params/M Size/MB FLOPs/G mAP@0.5/% FPS/
（frame‧s-1）
YOLOv8n（Baseline） 88.7 78.4 3.01 6.2 8.2 85.1 120 Baseline+GhostNet 91.1 76.6 2.30 4.9 6.3 82.9 128 Baseline+ShuffleNet V2 82.5 72.7 1.84 3.9 5.1 78.8 153 Baseline+EfficientNet V2 88.7 71 2.50 5.3 2.7 79.9 125 Baseline+MobileNet V3 85.4 75.5 1.19 2.7 2.8 81.7 131 Baseline+MobileNet V3_ECA 88.6 75 0.74 1.7 2.5 82.3 149 Baseline+MobileNet V3_CA 86.7 73.2 0.83 2.0 2.6 81.1 133 Baseline+MobileNet V3_CAA 87.1 78.9 2.61 5.6 4.8 83.5 121 Baseline+MobileNet V3_EMA 89.6 73.5 0.88 2.0 3.9 81.5 140
Table 4. Comparison of attention mechanisms with experimental results
View tableView in ArticleTable 4. Comparison of attention mechanisms with experimental results
Models P/% R/% Params/M Size/MB FLOPs/G mAP@0.5/% YOLOv8n+MobileNet V3_ECA（Baseline） 88.6 75 0.74 1.7 2.5 82.3 Baseline+GAM 86.7 73.2 0.75 1.8 2.5 81.1 Baseline+CA 89.7 73.1 0.74 1.8 2.5 82.5 Baseline+CBAM 89.1 78.6 2.61 5.6 4.8 83.3 Baseline+SE 89.6 73.5 0.74 1.8 2.5 81.5 Baseline+EMA 87.7 76.6 0.74 1.8 2.5 83.4 Baseline+CAA 82.5 79.9 0.74 1.8 2.5 84.1
Table 5. Experimental results of the loss function
View tableView in ArticleTable 5. Experimental results of the loss function
Models P/% R/% Param/M mAP@0.5/% CIoU 88.7 78.4 3.01 85.1 SIoU 88.6 78 3.01 84.5 WIoU v1 90.5 78.1 3.01 84.7 WIoU v2 89.3 76.8 3.01 84.9 WIoU v3 87.7 79.3 3.01 85.6
Table 6. Ablation experiments
View tableView in ArticleTable 6. Ablation experiments
Models Baseline MobileNet V3_ECA CAA ODConv WIoU P/% R/% Params/M FLOPs/G Size/MB mAP@0.5
/%
FPS/
（frame‧s-1）
Baseline √ 88.7 78.4 3.01 8.2 6.2 85.1 120 A √ √ 88.6 75 0.74 2.5 1.7 82.3 149 B √ √ √ 82.5 79.9 0.74 2.5 1.8 84.1 118 C √ √ √ 88.6 78 0.88 2.4 2.0 84.5 121 D √ √ √ √ 87.1 80.8 0.88 2.4 2.1 86 131 Ours √ √ √ √ √ 87.8 80 0.88 2.4 2.1 86.2 134
Table 7. Comparison between the original model and the improved model for the detection of various types of garbage
View tableView in ArticleTable 7. Comparison between the original model and the improved model for the detection of various types of garbage
Category YOLOv8n Improved-YOLOv8n P/% R/% mAP@0.5/% P/% R/% mAP@0.5/% Lunch boxes 73.4 64.3 69.3 72.4 65.7 72.8↑ Can 92.1 81 86.7 88.8 81 88.1↑ Peel 88.9 81.6 88.2 90.9 81.5 88.3↑ Garbage bags 91.5 71.8 83.5 92.4 74.4 80.8 Butt 93.6 80.3 84.9 97.4 78.2 87↑ Plastic bottles 94.9 83.2 91.9 93.3 83.5 92↑ Caps 87.3 66.7 74.7 87.2 71.6 79.4↑ Plastic bags 83 92.3 93.2 73 92.3 93.3↑ Carton 93.5 84.2 93.3 94.4 92.1 94.4↑
Table 8. Comparison of experimental results of different models
View tableView in ArticleTable 8. Comparison of experimental results of different models
Models P/% R/% Params/M Size/MB FLOPs/G mAP@0.5/% YOLOv5s 87.7 79.3 7.04 14.4 15.8 85.6 YOLOv5n 85.6 80.2 1.77 3.8 4.3 85.1 YOLOv10n 92.8 80.9 2.7 5.6 8.2 85.3 YOLOv7-tiny 86.1 79.2 6.03 12.3 13.2 84.7 Faster R-CNN 47.3 79.4 137.1 108.5 370.2 74.6 SSD 82.4 71.9 26.3 94.7 62.7 76.9 YOLOv8s 90.8 75.5 11.1 22.6 28.7 85.2 YOLOv8n 88.7 78.4 3.01 6.2 8.1 85.1 Ours 87.8 80 0.88 2.1 2.4 86.2
Tools
Get Citation
Copy Citation Text
Shizheng SUN, Lingling HE, Shuai ZHENG, Zeyin HE. Improved lightweight garbage detection method for YOLOv8n in complex environments[J]. Optics and Precision Engineering, 2025, 33(12): 1984
Paper Information
Category:
Received: Dec. 4, 2024
Accepted: --
Published Online: Aug. 15, 2025
The Author Email: Shizheng SUN (ssz091011@163.com)
© Copyright 2018-2021 | Chinese Laser Press.
All Rights Reserved 沪ICP备15018463号-20