Analyzing the complex relationships between surface environmental factors such as slope, aspect, precipitation, and land surface temperature and soil moisture in mountainous terrain is critical for achieving high-precision soil moisture inversion using optical and radar satellites. A representation learning-based convolutional neural network for mountain environments was proposed. Using domestic GF-1 remote sensing data and other datasets, the relationships between soil moisture and topographic factors in Chongqing central region were investigated. The convolutional neural network extracted mountainous environmental features and integrated high-accuracy soil moisture inversion data with multiple surface environmental factors. The results indicate that soil moisture has the strongest correlations with elevation and slope, with correlation coefficients of 0.70 and 0.69, followed by aspect, precipitation, and distance to rivers, with pronounced regional differences. These findings reveal interaction mechanisms between soil moisture and complex surface environmental factors in mountainous urban areas and support high-precision soil moisture remote sensing inversion models.