High-quality strong lens candidates in the final Kilo Degree survey footprint

We present 97 new high-quality strong lensing candidates found in the final∼350deg2, that completed the full ∼1350deg2 area ofthe Kilo-Degree Survey (KiDS). Together with our previous findings, the finallist of high-quality candidates from KiDS sums up to 268 systems. The newsample is assembled using a new Convolutional Neural Network (CNN) classifierapplied to r-band (best seeing) and g, r, i color-composited imagesseparately. This optimizes the complementarity of the morphology and colorinformation on the identification of strong lensing candidates. We apply thenew classifiers to a sample of luminous red galaxies (LRGs) and a sample ofbright galaxies (BGs) and select candidates that received a high probability tobe a lens from the CNN (PCNN). In particular, setting PCNN>0.8 for the LRGs, the 1-band CNN predicts 1213 candidates, while the3-band classifier yields 1299 candidates, with only ∼30\% overlap. Forthe BGs, in order to minimize the false positives, we adopt a more conservativethreshold, PCNN>0.9, for both CNN classifiers. This results in 3740newly selected objects. The candidates from the two samples are visuallyinspected by 7 co-authors to finally select 97 "high-quality" lens candidateswhich received mean scores larger than 6 (on a scale from 0 to 10). We finallydiscuss the effect of the seeing on the accuracy of CNN classification andpossible avenues to increase the efficiency of multi-band classifiers, inpreparation of next-generation surveys from ground and space.