Remote sensing image target detection method based on random access memory

Abstract

The invention discloses a remote sensing image target detection method based on random access memory, which is based on SSD network, maximum posteriori distribution, Fisher information, Laplace approximation and other methods. The method comprises the following specific steps: 1, reading image data and preprocessing the data; 2, constructing a convolutional neural network based on a fusion featurepyramid; 3, training the convolutional neural network to obtain static model parameters; 4, proposing a random access memory idea; and 5, detecting a remote sensing image target, and dynamically updating model parameters. According to the invention, a remote sensing image multi-class target detection algorithm under a unified framework is proposed to adapt to detection of the remote sensing imagetarget, and a random access memory thought is proposed by using maximum posteriori distribution to dynamically adjust model parameters in a detection stage, so that the model has good adaptive capacity and generalization capacity to new data. Remote sensing images are input, target category and position information are output, the automation degree is high, the efficiency is greatly improved, andthe cost is reduced.

Description

technical field [0001] The invention relates to a remote sensing image target detection method based on random access memory, in particular to a high-resolution visible light remote sensing image based on maximum a posteriori distribution MAP (Maximum A Posteriori) and SSD (Single Shot Multibox Detector) network in deep learning A deep learning method for target detection belongs to the technical field of high-resolution remote sensing image target detection. Background technique [0002] The science and technology of Remote Sensing originated from the unrecorded ground remote sensing in the 17th century. With the development of science and technology and the progress of the times, it has now developed into various types of remote sensors, multiple platforms, multiple system components, and multiple applications. Advanced and practical detection technology for the environment. As an important component of remote sensing technology, the detection and recognition technology o...

AI Technical Summary

Problems solved by technology

In practical engineering applications, the acquisition of training data is often done by non-uniform sampling in the entire data domain space, because the time domain and space domain for generating remote sensing image data are wider, and the non-uniform distribution of the collected training data is more obvious. The obtained parameter model cannot fully fit the entire data space, so the detection effect of the deep network model in the actual engineering will be good and bad. In severe cases, the model needs to be retrained and adjusted, which is time-consuming and labor-intensive. Difficult to meet the requirements of engineering applications

Images