Euclid: Identification of asteroid streaks in simulated images using deep learning

Open Access

Table 1

CNN results for non-centered cutout training, validation, and test sets.

Parameter	Training	Validation	Test
Accuracy	93.3%	93.2%	92.9%
Purity (precision)	97.6%	97.7%	97.0%
Completeness (recall)	61.4%	60.9%	59.3%
True positives	145 933	16 094	26 034
True negatives	1 185 564	131 738	218 531
False positives	3536	372	794
False negatives	91 873	10 342	17 831
Median coordinate error	0.135	0.134	0.138
Mean coordinate error	0.376	0.384	0.401

Notes. The training, validation, and test sets contain five times more negative training examples than positive ones. The units for coordinate errors are pixels.

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.