Research on Floc Feature Detection Method Based on Improved Density Map and Local Enhanced CNN

Fig. 9. Multi-point labeling and segmentation methods annotation information visualization. (a) Original image; (b) multi-point markup visualization; (c) image tags segment visual images

Download full size

Fig. 10. Binarization images obtained by different floc feature detection methods. (a) Sample floc image; (b) manual labeling; (c) GMM; (d) K-means; (e) Otsu threshold; (f) proposed method

Download full size

Table 1. Comparison results of the different methods
View table
Table 1. Comparison results of the different methods
Method $f_{M A E}$ $f_{M S E}$
CSRNet^［21］ 12.8 16.0
PDD-CNN^［22］ 13.6 17.7
Method in Ref.［23］ 12.5 15.3
DA-DCCNN^［24］ 13.2 17.3
Method in Ref.［25］ 12.9 16.9
Proposed 11.6 14.8

Table 2. Ablation study of LECNN
View table
Table 2. Ablation study of LECNN
Sequence Component $f_{M A E}$ $f_{M S E}$
1 VGG+Average 17.7 22.1
2 VGG+Gaussian 18.3 22.7
3 VGG+Average+1LEM 15.1 18.8
4 VGG+Average+2LEM 13.2 16.0
5 VGG+Average+3LEM 12.2 15.1
6 VGG+Average+3LEM+Residual 11.6 14.8

Table 3. Annotation information of multi-point labeling and segmentation methods
View table
Table 3. Annotation information of multi-point labeling and segmentation methods
Sequence $N_{m u t i_d e n}$ $T_{m u t i_d e n} / m i n$ $D_{s e g}$ $T_{s e g} / m i n$
1 152 1.25 986 11.0
2 151 1.40 975 12.3
3 144 1.20 856 10.0
Avg 149 1.28 952 11.1

Table 4. Comparison of proposed method with image segmentation-based floc feature detection methods
View table
Table 4. Comparison of proposed method with image segmentation-based floc feature detection methods
Method Sample Number Density Size $f_{M A E}$
GT 1 49 843 17.2
GMM 52 489 9.4
K-means 52 711 13.7
Otsu 52 740 14.2
Proposed 48 919 19.1 16
GT 2 60 753 12.55
GMM 58 385 6.64
K-means 62 613 9.89
Otsu 59 608 10.3
Proposed 57 788 13.8 6

Tools

Get Citation

Copy Citation Text

Jie Luo, Junran Zhang. Research on Floc Feature Detection Method Based on Improved Density Map and Local Enhanced CNN[J]. Laser & Optoelectronics Progress, 2025, 62(8): 0837005

Download Citation

EndNote(RIS)BibTex Plain Text

Set citation alerts for article

Save article for my favorites