Remote sensing image small target detection method based on improved YOLOv3

Abstract

The invention discloses a remote sensing image small target detection method based on improved YOLOv3, and belongs to the technical field of deep learning and target detection. The method comprises the steps: preprocessing a data set; optimizing a YOLOv3 network, and adding a cavity convolution group module, a feature enhancement module and a channel attention mechanism module in Neck; carrying out online data enhancement; performing forward reasoning; improving a loss function; and selecting the YOLOv3 network model with the highest detection precision and recall rate on the verification set to load the network, and the like. According to the method, the loss function is improved, and the cavity convolution group module, the feature enhancement module and the channel attention mechanism module are added into the YOLOv3 original network to improve the YOLOv3 detection network, so that the performance is obviously improved, the target detection in the remote sensing image is more comprehensive and higher in precision, and the training speed and the overall detection precision are improved.

Description

technical field [0001] The invention relates to the field of deep learning and target detection, in particular to a method for detecting small targets in remote sensing images based on improved YOLOv3. Background technique [0002] With the development of deep learning and neural networks, computer vision has developed rapidly. In this field, target detection and recognition technology has been widely studied and applied in practice, which brings great convenience to people's life. For example, when applied to drones, it can automatically identify specific targets in remote sensing images, and can replace manual work to efficiently complete this repetitive work. However, the following problems exist in many target detection work: [0003] 1. Most of the targets are small in scale, with only dozens of pixels, which is not conducive to finding and identifying; [0004] 2. The background is complex and there are many interference factors, such as shooting angle, lighting cha...

AI Technical Summary

Problems solved by technology

[0003] 1. Most of the targets are small in scale, with only dozens of pixels, which is not conducive to finding and identifying;

[0004] 2. The background is complex and there are many interference factors, such as shooting angle, lighting changes, similar targets, object occlusion, etc., which can easily lead to misjudgment and is not conducive to detection

Using this method to calculate the loss cannot fully reflect the positional relationship between the two frames, the coincidence degree, and the direction in which the predicted frames need to move closer, so the loss function needs to be improved

Images