Lightweight target detection method, apparatus and device for satellite-borne remote sensing image, and medium

Abstract

The invention provides a lightweight target detection method for satellite-borne remote sensing images, and the method comprises the steps: obtaining a first training sample set which comprises historical satellite-borne remote sensing images; training model parameters of the initial detection model by adopting a first training sample set to learn features of the satellite-borne remote sensing image to obtain a first detection model; wherein in the training process, the reciprocal of the model parameters is used as a constraint condition to constrain the distribution of the model parameters; obtaining a representative sample set from the first training sample set, and quantifying model parameters of the first detection model by using the representative sample set to obtain a second detection model; and inputting a to-be-detected satellite-borne remote sensing image into the second detection model to output a target detection result. The invention further provides a lightweight target detection device for the satellite-borne remote sensing image, electronic equipment and a computer readable storage medium.

Description

technical field [0001] The invention relates to the technical field of image processing and can be used in the technical field of spaceborne remote sensing image detection, in particular to a method, device, equipment and medium for light-weight target detection of spaceborne remote sensing images. Background technique [0002] Satellite data on-orbit processing refers to a working mode that relies on various computing devices onboard the satellite platform to process the data collected by the sensor in real time, and download the processing results as needed. This mode avoids large-scale satellite-to-ground data transmission and improves the agility of the system. Taking target detection as an example, compared with downloading the original image data, the amount of data downloaded from the on-board detection results is reduced by 2-3 orders of magnitude, with obvious advantages. . [0003] However, it is relatively difficult to deploy large-scale target detection algorith...

AI Technical Summary

Problems solved by technology

The above quantization effect is equivalent to setting a large number of parameters in the floating-point model to zero, which will reduce the fitting ability of the detection model and affect the accuracy of target detection.

Moreover, since the quantization parameters depend on the representative sample set, for data that is different from the set data distribution, the fitting ability of the quantized model will decrease, which will also affect the accuracy of target detection.

Images