Smart Agriculture

Figure/Table detail

Lightweight YOLOv8s-Based Strawberry Plug Seedling Grading Detection and Localization via Channel Pruning

CHEN Junlin, ZHAO Peng, CAO Xianlin, NING Jifeng, YANG Shuqin

Smart Agriculture, 2024, 6(6): 132-143. DOI: 10.12133/j.smartag.SA202408001

Model	P/%	R/%	mAP/%	F₁/%	FLOPs/G	Params/M	Model size/MB	Inference time/s
YOLOv8s	91.5	93.9	95.4	93.4	28.4	11.13	21.5	6.3×10^-3
Pruned-YOLOv8s	54.5	73.2	66.7	62.5	3.9	0.51	1.2	4.3×10^-3
Fine-tuned-YOLOv8s	91.9	95.1	96.4	93.5	3.9	0.51	1.2	4.3×10^-3

Table 4 Comparison of model performance before and after channel pruning for strawberry plug seedling-hole detection.

Other figure/table from this article

Fig.1 The research area and images of strawberry cutting seedlings
Fig.4 Strawberry plug seedling-hole detection model based on YOLOv8s
Fig.5 Channel pruning process based on LAMP Score
Fig.6 Flowchart of the two-stage strawberry plug seedling-hole matching and localization algorithm
Fig.9 Comparison of visualization results for different models in strawberry plug seedling and hole detection
Table 3 Experimental results of the two-stage strawberry plug seedling-hole matching and localization algorithm in multi-level strawberry plug seedling positioning research
Table 2 Parameter settings for channel pruning in the strawberry seedling-hole detection model
Table 1 Comparison of detection results for strawberry plug seedlings and holes using different network models
Fig.10 Detection and matching localization results of strawberry plug seedlings with different levels of invasion
Fig.11 Detection and localization failure results of the proposed algorithm
Fig.3 Annotation methods for various levels of strawberry plug seedlings and holes
Fig.2 Different levels of strawberry plug seedling image
Fig.8 Comparison of mAP and model size for the strawberry plug seedling-hole detection model under different pruning rates
Fig.7 The process of a two-stage matching and localization algorithm for strawberry plug seedlings and holes