Euclid Quick Data Release (Q1) XLI. AgileLens: A scalable CNN-based pipeline for strong gravitational lens identification

We present an end-to-end, iterative pipeline for efficient identification of strong galaxy-galaxy lensing systems, applied to theEuclidQ1 imaging data. Starting from the VIS catalogues, we rejected point sources, applied a magnitude cut (I-E <= 24) on the deflectors, and ran a pixel-level artefact-and-noise filter to build 96 & times; 96 pixel cutouts. The VIS+NISP colour composites were constructed with a VIS-anchored luminance scheme that preserves VIS morphology and NISP colour contrast. A VIS-only seed classifier supplies clear positives and typical impostors, which we used to curate a morphology-balanced negative set and augment scarce positives. Among the six compact convolutional neural networks (CNNs) that had been studied initially, the modified VGG16 (GlobalAveragePooling + 256/128 dense layers with the last nine layers trainable) exhibited the best performance. In our run, the training set grew from 27 seed lenses (augmented 67 & times; to 1809) plus 2000 negatives to a colour dataset of 30 686 images. After three rounds of iterative fine-tuning, a human grading of the top 4000 candidates ranked by the final model yielded 441 Grade A/B candidate lensing systems, including 311 overlapping with the existing Q1 strong-lens catalogue, and 130 additional A/B candidates (9 As and 121 Bs) not previously reported. Independently, the model recovered 740 out of 905 (81.8%) candidate Q1 lenses within its top 20 000 predictions, considering off-centred samples. Candidates spanI(E) similar or equal to 17-24 AB mag (median 21.3 AB mag) and are redder inY(E) - H(E)than the parent population, consistent with massive early-type deflectors. Each training iteration required about a week for a small team and the approach can easily be scaled to future wide-areaEuclidreleases. Subsequent works will focus on calibrating the selection function via lens injection, extending recall through uncertainty-aware active learning, and exploring multi-scale or attention-based neural networks with fast post hoc vetters that incorporate lens models into the classification.