Aiming at the problem that it is difficult to accurately segment the chest lung field affected by the lung shoulder area, thoracic diaphragm angle and ribs, we propose a lung field segmentation algorithm based on improved U-Net. First, the inception module is used to replace the convolutional layer in the U-Net coding block, which can increase the network width while capturing more image features. Then, the residual network is introduced in the coding block and the decoding block to increase the depth of the network and ensure the stability of the network. Skip connections are used between encoding and decoding to enhance the transfer and utilization of features, and to solve the problem of the loss of chest and lung field features due to continuous downsampling in the encoding part. Finally, the channel and spatial attention mechanism are combined in the encoding and decoding parts to analyze the image. Features are re-calibrated to effectively improve the segmentation accuracy of the algorithm. The experimental results show that compared with other segmentation algorithms, the segmentation performance of this algorithm is better. The accuracy, recall rate, specificity, and average intersection ratio on the public Montgomery County data set are 98.90%, 97.81%, 99.28%, and 97.17%, respectively.