Small Target Detection And Recognition Scheme For Water Surface Based On Hyperspectral Imagery

Target detection and classification recognition based on hyperspectral imaging occupy an important position in the field of remote sensing research.Due to the introduction of multi-dimensional features,the detection efficiency and recognition accuracy of Target detection and recognition schemes are low.Considering these two shortcomings above,under the premise of full understanding of hyperspectral image data,this paper studied the detection and recognition of small water targets in hyperspectral images and then suggested three key technologies,as follows:Firstly,the characteristics of the hyperspectral water index and the morphological characteristics of the small target are studied,and a small target detection scheme based on background priori is proposed.The Water Feature is used to accurately extract the background area of the water,and the small-object morphological features are used for highlighting the target of interest.Secondly,combined with the spectral enveloping line,a comprehensive detection method of background and target priori features is proposed.By extracting the envelope points and their corresponding valley values,the spectral slope characteristics and difference characteristics based on the enveloping line are calculated.The above characteristics reflect the changing trend and difference characteristics of the spectral lines respectively,and then use the synthetic features for pixel similarity measures.Finally,combining the characteristics of hyperspectral spectral lines with RNN,a deep-learning-based spectral line classification and recognition scheme is designed.Spectral rearrangement is used to raise the discrimination of spectral lines;Principal component analysis algorithm is used to reduce the dimensions of rearranged spectral data,which improves processing efficiency and reduces the influence of noise;combined with the advantages of RNN to process sequence data,high-precision recognition of target pixels is achieved.Comparison experiments with various methods show that the detection and recognition methods in this paper can detect small water targets completely,efficiently and accurately,and have strong anti-jamming capability for complex backgrounds.