Application of Image Segmentation Based on Deep Learning in Mobile Terminal Equipment

Currently, ultrasound image segmentation on mobile terminal has two limitations. First, the images collected by portable ultrasonic devices are of low quality and poor quality, which makes it difficult for traditional digital image processing methods to segment the target region accurately and stably. Second, the limited resources of mobile terminals make it difficult for deep learning models to operate efficiently on mobile terminals. Therefore, this paper improves the image segmentation model and applies it to portable ultrasonic equipment after compression. Firstly, the MobileNet V2 structure was used to replace the deep network of feature extraction in the Deeplabv3 + network model, so as to make it lightweight and reduce the parameters and calculation amount. Then, the model after training was quantified and compressed. Finally, the compressed model was applied to the neck artery image segmentation in portable ultrasound equipment. The direction of blood vessel and the Angle of line deflection are calculated from the segmenting images, and the automatic adjustment of color Doppler sampling frame is realized by using the parameters. Experimental results show that the improved deep learning model can accurately and quickly segment blood vessels on mobile devices.