Real-time Detection And Recognition Algorithm Of Hyperspectral Targets On Ocean

With the development of military information technology and science and technology,we urgently need to improve the perception of the ocean,especially the realtime detection and accurate identification of maritime targets.Hyperspectral imagery provides a large number of high-precision observation data for maritime target detection.Compared with SAR,hyperspectral image content is richer and more intuitive,and the acquisition method is simple and has irreplaceable advantages.Because hyperspectral images have the characteristics of uniform image and continuous imaging,hyperspectral imaging technology has great advantages in target detection and recognition in complex and wide-area marine environments.Small targets or camouflage hidden targets in remote sensing images are often difficult to detect through spatial features such as textures and edges,while hyperspectral images containing rich spectral information provide great convenience for detecting such targets.Target detection and recognition in the marine environment is an important issue in airborne remote sensing image processing.The traditional maritime target detection technology has a high false detection rate and missed detection rate due to the influence of sea clutter and seawater movement.At the same time,with the development of imaging spectroscopy,hyperspectral images have a higher spectral resolution and spatial resolution,while the amount of data has increased,and massive data has brought pressure to satellite downlink transmission,data storage and later data processing.Aiming at the problem that the traditional technology has high false detection rate and can’t detect in real time,this paper proposes a real-time detection algorithm for hyperspectral maritime targets and a hyperspectral image recognition algorithm based on detection results by using the advantages of hyperspectral target detection.Based on the practical application,this paper proposes a real-time target detection algorithm based on Gaussian mixture model and a hyperspectral image recognition algorithm based on convolutional neural network.The main work of the thesis is as follows:(1)For the target detection problem of massive hyperspectral data,this paper proposes a real-time detection algorithm for target processing.With the progressive realtime transmission of hyperspectral data,the algorithm only utilizes historical pixel information,and does not require unobtained pixel information,so that data transmission and processing are synchronized.At the same time,the Gaussian mixture model background modeling idea in the field of computer vision is applied.The Gaussian mixture model is used to model the background and update the ocean background model.The algorithm only needs the state of the previous moment and the current pixel information to update the current background model.Recalculating historical cells and storing all pixels reduces the running time of the algorithm while greatly reducing storage space.The main idea of the algorithm is to use the mixed Gaussian model idea to carry out background modeling;to process the data transmitted in real time,if the data matches the model,the background model is updated according to the current data;if it does not match,the object target is detected.Finally,based on the proposed real-time detection algorithm,this paper designs experimental data based on synthetic data and real hyperspectral data.The experimental results show that the real-time detection algorithm based on Gaussian mixture model background modeling realizes the real-time processing of the algorithm under the premise of ensuring the detection accuracy.(2)Target recognition is one of the important directions in the field of hyperspectral image processing,especially for the target recognition of high-spectral images at sea.Since most of the hyperspectral data at sea is seawater or cloud background,the target features appear in small probability.Therefore,the target-sensitive recognition algorithm becomes a key issue for offshore hyperspectral target recognition.In this paper,a hyperspectral image classification method based on convolutional neural network is proposed.The effectiveness of the proposed method is demonstrated by the labeled training data and the acquisition of raw data by the Xi’an Institute of Optics and Fine Mechanics,Chinese Academy of Sciences.The convolutional neural network proposed in this paper adopts the structure of convolutional layer,pooling layer and fully connected layer,effectively extracts the feature information of the maritime target,and maps the feature vector to the mark space through the fully connected layer to obtain a good high spectrum.Image classification results.