Botnet attacks can cause serious hazards such as data leakage, privacy compromise and malicious mining, seriously affecting the normal productive activities of network users. Due to the many types of botnet attacks, rich means and fast update rate of variants, botnet attack detection becomes a daunting challenge. Deep learning methods have the advantages of autonomous learning of traffic features and strong scene adaptation ability. Using deep learning methods for botnet attack detection can effectively improve the accuracy of botnet attack detection. In this paper, ECANet is introduced based on the ResNet model to propose a model EC-ResNet for detecting botnet attacks and improving the loss function during training to achieve high accuracy identification of botnet attacks. Experiments show that the detection accuracy of the EC-ResNet model proposed in this paper for botnet attacks is 93.45%, which is 1.59% better than that of ResNet.