Drone-Based Inventory Verification and Audit Quality: A Computer Vision Approach to Detecting Earnings Management through Physical Asset Verification

Abstract

This study proposes a two-stage methodology for using drone-captured inspection imagery as an audit-quality signal in detecting earnings management. The study integrates a computer-vision pipeline that verifies physical inventory with a financial-statement analysis grounded in the discretionary accruals (DACC) literature. Design. Stage 1 applies a pretrained YOLOv8 object￾detection model to oblique low-altitude drone imagery of vehicle storage facilities to produce an Inventory Verification Discrepancy Score (IVDS). Stage 2 estimates Modified-Jones discretionary accruals on a panel of 281 firms (1,048 firm-year observations, 2011–2022) and tests how IVDS relates to DACC, with inventory intensity as a theoretical moderator. Robustness checks include firm and year fixed effects and the exclusion of financial firms. Findings. YOLOv8n achieves 95.5% mean vehicle-detection accuracy on oblique drone imagery without domain fine-tuning, establishing the feasibility of off-the-shelf models for inventory verification in audit contexts. In the financial panel, DACC is significantly higher in inventory-intensive firms (t = 6.02, p < 0.001) and in non-Big4-audited firms (t = −3.01, p = 0.003) at the univariate level. Multivariate regressions reveal that, after controlling for inventory intensity, the IVDS × HighInv interaction is positive and marginally significant (p < 0.10), providing tentative support for the moderating role of inventory intensity. Firm- and year- fixed-effects specifications confirm a significant Big4 effect (β = −0.008, p < 0.05) that is otherwise absorbed by industry and size controls. Contribution. To our knowledge this is the first study to specify a complete pipeline linking drone-based asset verification to accrual-based earnings management. We provide a methodological foundation for integrating unstructured visual audit evidence with traditional financial-statement analysis, together with empirical evidence on the feasibility of off-the-shelf computer-vision models for low-altitude oblique drone deployment scenarios.