Lightweight SAR image ship detection model and method based on strip pruning

Abstract

The invention relates to a lightweight SAR image ship detection method based on stripe pruning, and belongs to the technical field of image processing. Extracting feature maps of different depths by using a strip convolution-based lightweight residual convolutional network; fusing different depth features by using a jump connection feature pyramid network to obtain fused depth feature maps of different scales; and predicting the probability of the corresponding position of each feature point and the vertical distance from the position of the feature point to the edge of the target by using a regression model based on the key points, and finally obtaining the position and probability of the target. According to the method, the lightweight residual convolutional network is obtained through one-time end-to-end training by using the lightweight residual convolutional network based on strip convolution, so that the training efficiency and the model robustness of the model are improved; the dimensionality of pruning is increased through strip pruning, and the parameter quantity and the calculated quantity of the model are further compressed.

Description

technical field [0001] The invention belongs to the technical field of image processing, and relates to a ship target detection method of a SAR image, in particular to a lightweight SAR image ship detection method for limited storage and computing equipment. Background technique [0002] Because of the all-day and all-weather imaging characteristics of SAR images, the ship detection of SAR images is of great significance in the fields of marine monitoring and military affairs. However, limited by the hardware conditions of the spaceborne terminal environment, the SAR image ship detection method needs to reduce the parameter scale and calculation amount as much as possible under the premise of ensuring the detection performance. [0003] "On-Board Real-Time Ship Detection in HISEA-1 SAR Images Based on CFAR and Lightweight Deep Learning" (Remote Sensing.IEEE, 2021:1995-2013.) uses a two-parameter constant false alarm rate to extract images that may contain ships, Then use th...

AI Technical Summary

Problems solved by technology

This method is independent of pruning and training process, and cannot achieve end-to-end training

It only performs pruning at the channel layer, and does not completely solve the problem of parameter redundancy

Images