Abstract: Accurate zoning of geological hazard susceptibility is a key and prerequisite for hazard prevention and reduction, as well as national spatial planning. This paper takes Ledu District as the research area, uses the “frequency ratio method” to determine the sensitivity of geological hazards in each grading interval of 9 influencing factors, uses the “variance inflation factor method” to determine the correlation of factors, establishes logistic regression and convolutional neural network models, and constructs a geological hazard susceptibility evaluation system that couples logistic regression and convolutional neural network models. The geological hazard susceptibility evaluation of Ledu District is completed and the model accuracy is compared. The results show that elevation has the most significant effect on geological hazards, followed by engineering rock formations, slopes, rivers, roads, TWI, faults, slope orientation, terrain undulation; the evaluation index and AUC value of LR-CNN coupling model are better than that of single model, and its AUC value is 0.25 and 0.12 higher than that of LR model and CNN model respectively; the highly susceptible areas are mainly distributed in loose and weak rock areas with elevations of 1 788-2 848 m, characterized by gentle slopes, low TWI values, small terrain undulations, and proximity to rivers and roads.