Sonar target detection method based on ssd

Abstract

The invention discloses an SSD-based sonar target detection method. The current underwater target detection and recognition methods are difficult to accurately identify multiple targets simultaneously. The present invention is as follows: 1. Establish the SSD neural network model. 2. Generating n pieces of sonar images to be tested by using the sonar data to be identified. 3. Send the measured sonar image to the SSD neural network model to obtain the feature map. 4. Set the detection frame on the feature map. 5. Input the detection frame into two convolution kernels to obtain the category score and shape offset of the detection frame for each target category. 6. Determine the target types contained in the measured sonar image, and select all targets. The present invention integrates deep learning technology into sonar target detection, inputs sonar images generated by sonar data into the SSD neural network model, and completes feature extraction, target detection, and target classification in the model at one time, thereby greatly improving the detection speed.

Description

technical field [0001] The invention belongs to the cross technical field of artificial intelligence and underwater acoustic electronic information, and specifically relates to a three-dimensional imaging sonar target detection method based on deep learning. Background technique [0002] With the continuous iteration and development of scientific and technological information technology, underwater detection technology has been greatly promoted in recent years, and has a wide range of application fields, such as small target defense in military, dangerous target screening, target tracking, etc.; It has important functions such as seabed resource exploration and measurement, tracking and protection of endangered organisms, and seabed modeling. [0003] Underwater target detection and recognition is an important part of modern sonar systems and underwater acoustic countermeasures. It is the research focus of each country's maritime security. It has been widely concerned by sch...

AI Technical Summary

Problems solved by technology

[0009] The contrast and signal-to-noise ratio are low, and they are greatly affected by noise. The current underwater target detection and recognition methods still have many bottlenecks in this case, such as incomplete or slow extraction of sonar image target features, small underwater targets due to contrast The low signal-to-noise ratio is falsely detected or missed, and high precision, strong robustness, and real-time performance of the system are taken into account at the same time. For other traditional deep learning target detection network models, the model is too large, the parameters are huge, the operating memory is high, and the speed Slow and narrow application range, this design uses the mobilenet network as the basic network for feature extraction, using a streamlined structure using depthwise separable convolutions (Depthwise Separable convolutions) to construct a light weight deep neural network, while ensuring accuracy Under the premise, the amount of model parameters is effectively reduced, the detection speed is accelerated, and the requirements of mobile applications are met.

Images