Image Classification Of Submairne Volcanic Smog Map Based On Convolution Neural Network

The formation and evolution of submarine volcanoes are closely related to ecosystems,and the detection of the existence of submarine volcanoes is of great significance to the development of modern science.The more common method of judging whether there is a volcano on the sea floor is to detect whether there is smoke on the sea floor.However,the existing smoke recognition methods have the problems of insufficient efficiency and insufficient effect,and mostly recognize the smoke produced on land,such as forest fire smoke,community fire smoke recognition and warning,etc.Relevant research on identification is relatively fragmented,and the effect is not satisfactory.The seabed environment is more complex than the land environment,and there is more noise interference.Common image processing algorithms such as feature extraction are difficult to achieve the purpose of seabed volcanic smoke recognition.In this paper,smoke recognition is combined with deep learning algorithms in the field of machine learning,and the convolutional neural network model is applied to submarine volcanic smoke recognition.Combined with existing research results and technologies,the volcanic smoke images in the deep sea submarine scene Recognition classification overcomes the shortcomings of other image recognition schemes,improves the efficiency of deep learning algorithms,and increases the accuracy of submarine volcanic smoke recognition.The technology of image classification using convolutional neural networks is currently widely used in various fields.Among them,the technology of image recognition is the key.How to improve the recognition rate and accuracy of recognition is of great significance to scientific research and is related to the image.The safety and practical effect of identification.The main content of this article is to use a variety of data enhancement techniques to expand the data set with limited computing power and a small amount of data set,and then apply the convolutional neural network to train the data set,so as to achieve a smoke map,no smoke Undersea maps are used to identify and classify,and then determine whether there is a possibility of submarine volcanoes.The flow of experimental work in this article is mainly:(1)To study and compare the effects of Harris corner algorithm,SIFT algorithm and Gabor wavelet algorithm on submarine volcanic smoke image-free processing,confirm that this experiment uses convolutional neural network model and Gabor wavelet feature extraction model for submarine volcanic smoke image processing.(2)Make the data set.First of all,the data set is collected through the network,and the collected deep-sea submarine pictures are divided into two types: smoked submarine pictures and smoke-free submarine pictures.A variety of data enhancement techniques are used,including angle rotation,horizontal flip,random cropping,and Gaussian noise.Expanded the data set,produced a total of 3348 positive samples and 3348 negative samples,and then unified the network input of all samples,and finally divided all the samples into two parts of the training set and the test set to complete the production of the data set.(3)Research on the structure of convolutional neural networks,build 3-layer and 4-layer convolutional neural networks,respectively,in terms of training models,selecting activation functions and adjusting parameters.Use the seabed volcanic smoke images collected and produced in the previous period to train the convolutional neural network model,adjust the parameters of the experiment according to the efficiency of the training process and the training effect,and improve the structure of the experimental convolutional neural network model.The three-layer and four-layer convolutional neural networks are compared in a series of indicators such as the convergence efficiency of the submarine volcanic smoke classification and the recognition accuracy of the Gabor wavelet feature extraction model.Adjust the parameters of the convolutional neural network according to the recognition efficiency and the results,and improve the model.By comparing the effect of 3 layers and 4 layers of convolutional neural network and Gabor wavelet feature extraction model on the identification of submarine smoke images,the results of this experiment show that: using 4 convolutional layers,4 maximum pooling layers,2 joins A four-layer convolutional neural network composed of Dropout's fully connected layer and a SoftMax classifier layer was used to identify the seabed volcanic smoke image.Its classification accuracy reached 94.33%,which was 5.16% higher than the three-layer convolutional neural network.Gabor wavelet feature extraction flue gas dynamic accumulation model improved by 5.66%.And the 4-layer convolutional neural network network training process converges faster,which is more suitable for recognition and classification of seabed volcanic smoke images.