In the context of the 3.5 GHz Citizens Broadband Radio Service (CBRS) band, accurate prediction of path loss is critical to mitigate potential interference risks for incumbent users from lower-tier users [1]. The current CBRS standards, as stipulated by the Wireless Innovation Forum, rely on the Longley- Rice model, an irregular terrain model (ITM), for path loss estimation [2]. However, this model lacks the incorporation of clutter data, leading to an underestimation of path loss. This study proposes an innovative approach that integrates deep learning (DL) techniques with model-assisted methodologies leveraging satellite imagery to enhance the accuracy of path loss predictions [3, 4]. Through numerical analysis, the effectiveness of this proposed method is demonstrated, achieving a root mean square error (RMSE) of 4.23 dB [5]. This performance surpasses that of the Longley-Rice model and various tuned or adapted propagation models. The integration of DL and satellite imagery proves to be a promising avenue for advancing path loss prediction capabilities in the 3.5 GHz CBRS spectrum range.