The detection accuracy of the gap in the internal threaded joint of a rocket engine is an important indicator of its quality. Due to the complex internal surface of the engine shell, the quality of the internal gap is not only low in efficiency but also in poor reliability by manual inspection. Proposing a visual inspection method for inward seams based on FNN network. The feature parameters of the image are constructed with gray-level co-occurrence matrix and PCA algorithm, and the FNN network is trained to classify and classify the rough and finished surfaces of the internal seams of rocket engine shells. The recognition rate is 98.8%; then, different image processing is performed for the two types of situations, and the Sobel operator is used to find the edge of the gap; finally, the system error of the algorithm (collecting the original image error, the straight line fitting error) is corrected through calibration, and the internal slit is completed Width is accurately measured. Experiments show that the method is stable and reliable, and can achieve a recognition accuracy of ±0.02mm in the range of 0.1mm-0.6mm. This method realizes the high-precision measurement of the threaded joints inside the rocket engine shell, and provides technical guarantee for the realization of high-efficiency automatic production and quality inspection of products.