This paper presents a broadband, low-noise optical edge detection system for incoherent infrared radiation, designed to offload low-level image feature extraction from deep neural networks in resource-limited applications. The system combines a 24-mm aperture metasurface with a refractive lens to achieve Laplacian-based edge detection across the 7.5- to 13.5-μm long-wave infrared (LWIR) imaging band. An inverse design approach is used to optimize the metasurface, allowing for easy integration with uncooled microbolometer-based LWIR imagers. The system demonstrates improved performance over conventional methods, including higher signal-to-noise ratio (SNR) and better edge detection in low-contrast scenes. Additionally, a polarization multiplexed approach using a birefringent metasurface is introduced for single-aperture implementation. This work could enhance computer vision capabilities in resource-constrained systems by leveraging optical preprocessing to reduce computational requirements for high-accuracy image segmentation and classification.This paper presents a broadband, low-noise optical edge detection system for incoherent infrared radiation, designed to offload low-level image feature extraction from deep neural networks in resource-limited applications. The system combines a 24-mm aperture metasurface with a refractive lens to achieve Laplacian-based edge detection across the 7.5- to 13.5-μm long-wave infrared (LWIR) imaging band. An inverse design approach is used to optimize the metasurface, allowing for easy integration with uncooled microbolometer-based LWIR imagers. The system demonstrates improved performance over conventional methods, including higher signal-to-noise ratio (SNR) and better edge detection in low-contrast scenes. Additionally, a polarization multiplexed approach using a birefringent metasurface is introduced for single-aperture implementation. This work could enhance computer vision capabilities in resource-constrained systems by leveraging optical preprocessing to reduce computational requirements for high-accuracy image segmentation and classification.