To addresses the challenge of distinguishing between malignant and benign tumors in breast cancer ultrasound images， an improved method based on the EfficientNet model is proposed. This thesis introduced an enhanced whale optimization algorithm (WOA) and a global context (GC) module to improve the accuracy and efficiency of early breast cancer detection. The model optimizes feature extraction and classification performance by combining depthwise separable convolution and large kernel convolution. Additionally, dynamic hyperparameter tuning and data augmentation were applied to further enhance the model′s generalization ability and stability. Experimental results show that the model achieved an accuracy of 99.81% on the training set and 98.06% on the validation set, significantly surpassing traditional methods. The mean average precision (mAP) was increased from 96.42% to 98.60%, demonstrating the model′s effectiveness in improving the accuracy and reliability of early diagnosis， providing an efficient technical pathway for early screening and diagnosis of breast cancer.