Abstract: The interference signal caused by weld surface texture leads to the difficulty of the defect detection in eddy current detection technology. A dual-frequency data fusion method combined with the design of the sensor was proposed. The sensor includes a rectangle excitation coil and a giant magneto resistance (GMR) element. 500 Hz and 4 kHz are chosen as the excitation frequencies, and the distribution of magnetic induction intensity at the excitation frequencies of 500 Hz, 1 kHz and 4 kHz were simulated by the finite element analysis. The affine transformation matrix of real part and imaginary part data at the two frequencies was constructed through the least square method. The detection affine transformation data at the reference frequency were fused with the data at the basic frequency. The experiment was carried out on aluminum-alloy welding defect. The result indicates that this method suppresses not only the interference signal of weld surface texture, but also the interference caused by the angle of two welded aluminum-alloy plate. The defects of less than 1 mm in diameter were hardly reflected though data at two excitation frequencies independently, but they can be successfully detected by the proposed method.