Stream automatic detection with convolutional neural networks

Open Access

Table 3

Comparison SAD-CNN: V-2 & V-3.

		SAD-CNN: Version-2			SAD-CNN: Version-3
Inclination	Metric	Precision	Recall	F1-score	Precision	Recall	F1-score	Support
	Not Det	0.84	0.76	0.80	0.81	0.81	0.81	42
15°	Det	0.86	0.91	0.89	0.88	0.88	0.88	68
15°	Average	0.85	0.84	0.84	0.85	0.85	0.85	110
	Accuracy			0.85			0.85	110

	Not Det	0.73	0.58	0.65	0.72	0.89	0.80	38
30°	Det	0.77	0.87	0.82	0.92	0.79	0.85	62
30°	Average	0.75	0.72	0.73	0.82	0.84	0.83	100
	Accuracy			0.76			0.83	100

	Not Det	1.00	0.30	0.46	0.84	0.74	0.78	42
45°	Det	0.66	1.00	0.80	0.84	0.90	0.87	63
45°	Average	0.83	0.65	0.63	0.84	0.82	0.83	105
	Accuracy			0.70			0.84	105

	Not Det	0.83	0.54	0.65	0.79	0.82	0.80	122
All	Det	0.76	0.93	0.83	0.88	0.86	0.87	193
All	Average	0.79	0.73	0.74	0.84	0.84	0.84	315
	Accuracy			0.77			0.84	315

Notes. Comparison of SAD-CNN metrics for Version-2 and Version-3 at different inclinations (15°, 30°, 45°) and overall (All). Metrics include Precision, Recall, F1-score, and Accuracy for each configuration. The metrics obtained for the test set (315 images) of Group III with a total of 5234 images at different inclinations. The values were obtained with a threshold value of 0.5.

Current usage metrics show cumulative count of Article Views (full-text article views including HTML views, PDF and ePub downloads, according to the available data) and Abstracts Views on Vision4Press platform.

Data correspond to usage on the plateform after 2015. The current usage metrics is available 48-96 hours after online publication and is updated daily on week days.

Initial download of the metrics may take a while.