Magnetic Anomaly Detection (MAD) is a key method in geophysical exploration, ferromagnetic target identification, and medical monitoring. Traditional methods based on mathematical modeling and signal processing face challenges in high-noise environments. Deep learning approaches, especially Convolutional Neural Networks (CNNs), have high capability in automatic feature extraction but require large labeled datasets. In this paper, a hybrid method based on one-dimensional convolutional neural networks (1D-CNN) and transfer learning is proposed for magnetic anomaly detection. First, using Poisson disk sampling in the position-angle space, fixed-length 200-sample linear profiles are extracted from the simulated magnetic field. Then, a deep autoencoder is trained on 22,508 synthetic one-dimensional signals to learn a compact signal representation. In the next stage, the encoder part of this network is used as a fixed feature extractor, and by adding classification layers, the final model is trained on 5,130 target profiles. Evaluation results show that the proposed method with 97% accuracy and F1-score of 97% achieves a significant improvement compared to the baseline model without transfer learning (90% accuracy). This approach provides an effective and reliable solution for practical applications with limited data.