The Study Of Ship Target Detection In Optical Remote Sensing Images

Ship target detection technology is a very important subject in remote sensing satellite image processing and analysis field.Especially for the high resolution optical remote sensing image,the massive data can provide more detailed information,but it seriously restricts the detection efficiency of the ship target.Therefore,how to get the position information of the ship target quickly and accurately has become a hotspot topic.In order to solve the problem of target detection in optical remote sensing image,this paper focuses on the extraction of ship target candidate area and the identification technology of ship target,which improves the accuracy and efficiency of ship target detection.The main work and contribution of this paper can be summarized as follows:Firstly,the section is mainly consisted of image filtering,uniform illumination and de-fog interference based on the preprocessing of optical remote sensing image,especially image uniform illumination and de-fog interference.Its purpose is to reduce the impact of noise,non-uniform illumination and fog,which makes the target information highlight.Secondly,the method of extracting candidate area of ship target is studied,and an improved method is proposed to extract the candidate area of ship target.In this paper,the PQFT method is introduced,and the wavelet transform is added to the original PQFT algorithm.The saliency features of the image are analyzed from different scales,and the image with the best salience is selected.The experimental results show that the improved PQFT method is better than the original algorithm,and the computing time is reduced.Thirdly,In order to determine whether the target of the extract candidate region is a ship target,the shape feature,gray feature,texture feature and histogram of oriented gradient(HOG)feature of the ship are extracted in this section.And an improved LBP feature extraction algorithm is proposed,which has stronger anti-interference ability and lower computational complexity,and increased the controllability of LBP feature extraction.By comprehensive information of the ship target in many aspects,these methods could determine accurately the ship target or false.At last,the improved Extreme Learning Machine(ELM)algorithm is used to classify the ship target.ELM is a neural network algorithm,which is characterized by a single hidden layer,the artificial setting of hidden layer nodes,the random generation of input weights and hidden layer bias.So that the method costs a short calculation time,the stability of generalization is better and it's not easy to fall into the local optimum.The activation function of traditional ELM algorithm,such as sigmoid function,sin function and tanh function,is over-saturated.In this paper,a non-linear rectified ELM algorithm is proposed.Finally,the improved ELM algorithm is applied to the classification of ship targets.Based on the above research,the MATLAB platform is used to validate the simulation results.The experimental results show the effectiveness of the improved algorithm and increase the detection precision and efficiency.