Research On Ship Target Recognition Based On Convolution Neural Network

Purpose — Convolution neural network,as a deep neural network structure,improves the training efficiency of convolution neural network on large-scale image data sets by gradually compressing the length and width of image data,expanding the number of channels of data,and continuously adjusting the adaptive ability of convolution kernel in the training process of reducing errors.Convolution neural network can generate a network model with high accuracy under normal circumstances and apply it to practical research.In rare cases,the corresponding model can be obtained by reducing the error several times,but this does not affect the performance of convolution neural network.From the final training results,the accuracy of the model and the stability of the algorithm are improved compared with the traditional neural network.In order to improve the recognition accuracy of warship targets,a 12-layer convolutional neural network model is first constructed.Then,in order to further improve the recognition accuracy,the 12-layer convolutional neural network and SVM algorithm are applied to warship target recognition.Then,the proposed 12-layer convolutional neural network is improved by using the global average pooling method.Finally,a 12-layer neural network based on full convolution is proposed to improve the recognition rate of warship images and the computational efficiency of the whole model.Method — Based on the analysis of ship target features,in order to overcome the defects of traditional methods in image preprocessing and recognition,convolution neural network is used to extract image features and classify the images.Through training with a large amount of ship image data,automatic feature extraction is realized.At the same time,ship target recognition based on convolution neural network model is formed by making it have stronger robustness and generalization ability through many iterations and reducing errors.Research results — Firstly,data sets are preprocessed by some methods of data enhancement and image noise reduction,and the proposed convolution neural network model is designed and optimized.The preprocessed image data set is sent to the convolution neural network for training,testing and optimization adjustment,and finally a naval vessel target recognition model based on the convolution neural network is formed,with high accuracy.Limitations of research — The number of images in the data set is limited.Convolution neural network can only identify the types of ships collected in the data set,and these types can only represent a small part of real life,and cannot identify the types of ships other than the data set.If more complex background environment is encountered in the actual process,whether it can be accurately identified still needs to be verified.Actual impact — After using data enhancement technology,the number of images in the data set has increased significantly.After using image noise reduction technology to process images,the ship features that people are interested in are more obvious,the outline of the ship is more prominent,and the noise on the original image is significantly reduced,thus increasing the accuracy rate.Originality — The method of data enhancement and image noise reduction is used to process the pictures in the data set,and the resolution of the image is improved by adjusting the gray values of pixel points,etc.The convolution neural network is used to recognize and classify the processed images,and then the proposed convolution neural network is improved to improve the recognition rate while reducing the calculation amount of the model and the number of parameters in the model.