Fog and smog often render scene details invisible, making computer-vision tasks like object detection and recognition difficult. Traditional dehazing methods rely on direct estimation of transmission maps and atmospheric light, which is both noise-sensitive and computationally costly. Dark Channel Prior (DCP) relieves these requirements by leveraging low-intensity patches to identify thick haze, but its performance degrades in heterogeneously hazed scenes. To improve its results, this paper proposes a Dark Channel Prior Infused All-in-One Dehazing Network (DCPI-AODNet). With the incorporation of DCP on an adapted All-in-One Dehazing Network (AOD-Net) in addition to the RGB input, the proposed network is able to learn more accurate transmission patterns, especially in complex scenes where haze displays complicated and non-homogeneous patterns. In contrast to traditional pipelines, the modified AOD-Net dispenses with independent atmospheric light and transmission inference, providing end-to-end dehazed results via a light-weight convolutional architecture conducive to real-time use. Combining DCP guidance maintains the statistical advantages of conventional priors with the data-driven flexibility of deep learning. Experimental assessments yield consistently sharper textures, more accurate colors, and reduced halo artifacts compared to baseline AOD-Net and single-matting DCP, showing enhanced visual quality under diverse fog and smog environments. Quantitative performance gains in peak signal-to-noise ratio and structural similarity also confirm these subjective improvements, making the model an effective solution for computer-vision tasks.