Lung cancer threatens human life and health with high morbidity and mortality. CT images of lung cancer are an important basis for early diagnosis. Based on the original CNN, this paper proposes a lightweight ResNet-CNN ensemble model, which uses elastic transformation data enhancement strategy, ResNet18 and CNN as learners, and gradient boosting tree as a classifier. Experiments show that the accuracy of the model is 98.32%, and the precision value, recall rate, and F1 score are all over 0.98, which are 5.35 and 0.93% higher than the original model and ResNet101, respectively, and the parameter quantity is only 84.38 MB, which is much lower than ResNet101’s 260.73 MB. The model in this paper achieves the goal of lightweight and high performance, and provides a new method for early diagnosis of lung cancer, reducing artificial error rate and reducing the burden of medical staff.