Alternating current field measurement (ACFM) technique has been widely used in defect inspection of underwater metal structures. However, due to the high cost of inspection, it is difficult to carry out nondestructive testing (NDT) regularly to obtain the health status of key nodes of structures. The quantification of defects depends on a simple empirical formula, which leads to large quantization errors. In this article, a flexible ACFM magnetic sensor array for in situ inspection of underwater structural cracks is proposed. An in situ inspection probe based on the sensor array is manufactured and can be fixed on the surface of underwater curved structures. The probe can realize visual imaging of cracks in underwater carbon steel structures to obtain the damage status of key nodes of the structure regularly. According to the inspection image, a crack quantification method based on multiphysical features fusion convolution neural network is used for the quantification of crack length, depth, and angle. The experimental results show that the in situ inspection probe can realize the visual imaging of cracks. The quantification method can accurately evaluate the crack dimensions. The average length error, average depth error, and average angle error are 0.41 mm, 0.169 mm, and 1.131 degrees, respectively.