SGAA (subgroup‑guided active alignment) uses image learning to correct automated optical subgroups that suffer from uncompensated aberrations. It collects multi‑field point spread functions (PSFs) from a special test setup, calibrates with a golden sample, and trains a single regression model to predict the correction needed for each subgroup to match a common optical target. In computer simulations that included realistic tolerances, one SGAA step cut the mean differential wavefront RMS from 0.0592 λ to 0.0086 λ – an 85.4 % improvement. The same approach was applied to 100 production subgroups. SGAA raised the full‑lens weighted modulation transfer function (WMTF) by 0.073, 0.065, and 0.040 at 30, 50, and 80 lp/mm, respectively, compared to traditional industrial MTF‑based scanning. Acceptance yield for an 8‑MP image rose from 52.5 % to 81.7 %, a 29.2‑percentage‑point gain. Meanwhile, the average alignment time dropped from 43 s to 3.1 s. The technique needs a dedicated test configuration to capture subgroup PSFs and a golden reference, which may limit its use to lens designs that can accommodate this setup.