InvenSight: AI-Powered Warehouse Damage Detection System Using ResNet-Based Classification

This study investigated the impact of five different optimizers—Adam, SGD with momentum, RMSProp, NAdam, and Adamax—on the convergence behavior and generalization of a ResNet-50-based vision inspection framework. Experimental results demonstrated that Adam consistently achieved the fastest convergence and highest validation accuracy (87.10%), while RMSProp and SGD with momentum also showed competitive performance. Fine-tuning the last 20 layers of ResNet-50 using the optimal optimizer further enhanced domain-specific feature adaptation, improving validation accuracy by approximately 2–3%. The resulting InvenSight system successfully transforms damage inspection from a human-bottlenecked, inconsistent, non-auditable process into an AI-first, human-verified, fully-logged, analytics-friendly workflow that scales with warehouse throughput rather than with headcount. The system is ready for real-world pilot deployment, and its modular architecture leaves ample room for enhancements including multi-class damage taxonomy, YOLO-based localization, conveyor diversion integration, and WMS plug-in adapters. Smarter Damage Control, Stronger Supply Chain.  

[1] V. Kumar and M. Kumari, "Barcode recognition using CNN for logistics automation," Int. J. Comput. Appl., 2018. [2] S. Lee and J. Yoon, "Automated parcel classification using convolutional networks," IEEE Access, 2019. [3] Y. Zhang and X. Li, "Surface defect detection in manufacturing using deep learning," IEEE Trans. Ind. Informatics, 2020. [4] K. He, X. Zhang, S. Ren, and J. Sun, "Deep residual learning for image recognition," in Proc. IEEE Conf. Comput. Vis. Pattern Recognit., 2016. [5] J. Deng et al., "ImageNet: A large-scale hierarchical image database," in Proc. IEEE Conf. Comput. Vis. Pattern Recognit., 2009. [6] J. Vater, M. Kirschning, and A. Knoll, "Real-time error detection in production using edge computing," in Proc. Int. Conf. Omni-Layer Intell. Syst., 2020. [7] G. Litjens et al., "Deep learning in medical image analysis," Medical Image Analysis, vol. 42, pp. 60–88, 2017. [8] J. Liu et al., "Steel-surface defect classification with deep residual learning," Computers in Industry, 2021. [9] Q. Li and Y. Huang, "Defect detection in semiconductor wafers using CNN," Sensors, 2020. [10] A. Saeed et al., "Multi-sensor data augmentation for robust sensing," in Proc. Int. Conf. Omni-Layer Intell. Syst., 2020.