Title:New high-quality strong lens candidates with deep learning in the Kilo Degree Survey

Abstract:We report new high-quality galaxy scale strong lens candidates found in the Kilo Degree Survey data release 4 using Machine Learning. We have developed a new Convolutional Neural Network (CNN) classifier to search for gravitational arcs, following the prescription by Petrillo et al. 2019 and using only $r-$band images. We have applied the CNN to two "predictive samples": a Luminous red galaxy (LRG) and a "bright galaxy" (BG) sample ($r<21$). We have found 286 new high probability candidates, 133 from the LRG sample and 153 from the BG sample. We have then ranked these candidates based on a value that combines the CNN likelihood to be a lens and the human score resulting from visual inspection (P-value) and we present here the highest 82 ranked candidates with P-values $\ge 0.5$. All these high-quality candidates have obvious arc or point-like features around the central red defector. Moreover, we define the best 26 objects, all with scores P-values $\ge 0.7$ as a "golden sample" of candidates. This sample is expected to contain very few false positives and thus it is suitable for follow-up observations. The new lens candidates come partially from the the more extended footprint adopted here with respect to the previous analyses, partially from a larger predictive sample (also including the BG sample). These results show that machine learning tools are very promising to find strong lenses in large surveys and more candidates that can be found by enlarging the predictive samples beyond the standard assumption of LRGs. In the future, we plan to apply our CNN to the data from next-generation surveys such as the Large Synoptic Survey Telescope, Euclid, and the Chinese Space Station Optical Survey.