Research On Forward Looking Sonar Image Classification Based On Deep Learning

In recent years,all countries are actively exploiting and utilizing marine resources.Because the propagation speed of underwater light is affected,sonar is a more effective method.Therefore,it is most suitable to use acoustic wave for underwater observation.With the development of underwater sonar technology,the research of image classification based on forward looking sonar is very necessary,which is of great significance to military and civil.For Underwater Unmanned equipment,sonar image detection and classification is an important deduction basis for decision optimization.However,marine debris also has a great impact on environmental pollution.Sonar image classification is helpful to capture seabed debris.The deep learning technology represented by convolutional neural network has a good effect in the field of computer vision,such as video classification.This paper proposes to use the depth neural network to classify the ocean debris image of forward-looking sonar.In this paper,the application of depth neural network in the classification of marine debris forward looking sonar images is comprehensively evaluated from the perspectives of accuracy and model complexity.There is no common marine debris data set in the forward-looking sonar image data set.The data set used in this paper is the data set captured in the water tank system Laboratory of Watt University in Edinburgh,Scotland.In this water tank system,sonar sensor is used to capture the sonar image of the data set.The sonar sensor used to capture data is the aris-Explorer-3000,a forward-looking sonar that is also known as an acoustic camera.Traditional forward-looking sonar image classification mainly uses template matching algorithm,which is a specific method to calculate the similarity between two images.If there is a set of labeled template images available,then by calculating the similarity with each template image,the most similar image is found,and the class label of the image is output,then the given test image can be classified.In this paper,we propose a new algorithm based on the neural network,which is based on the large scale of the underwater acoustic network Net fire module is redesigned.In this paper,firstly,the ocean debris data set of forward-looking sonar is preprocessed,image enhancement and image denoising are carried out,and the small data set is expanded to avoid over fitting.According to the characteristics of forward-looking sonar ocean debris image and the suppression of irrelevant details,a lightweight convolutional neural network at squeeze net based on attention mechanism is proposed.In order to reduce the parameters,After multi-scale feature fusion,the fire module is applied to the new network to solve the classification problem.The experimental results show that AT-squeeze-net have good classification performance,good convergence performance and high real-time requirements.