Deep learning small target detection method and device based on cascade fusion and attention mechanism

Abstract

The invention discloses a deep learning small target detection method and device based on cascade fusion and an attention mechanism, and the method comprises the following steps: S1, inputting a to-be-detected image, and carrying out the preprocessing of the to-be-detected image; s2, performing feature extraction on the preprocessed image by using a deep convolutional neural network based on cascade fusion and an attention mechanism, extracting to obtain target image features, performing feature fusion by a feature cascade fusion layer based on a cascade feature fusion structure, enabling the spatial attention mechanism layer to obtain a semantic mask of the small target area, fusing the semantic mask with the original features channel by channel, and outputing extracted target image features; and S3, carrying out prediction and post-processing on the extracted target image features to obtain a final target detection result, and outputting the final target detection result. The invention can achieve small target detection based on deep learning, and has the advantages of being simple in implementation method, low in cost, high in detection efficiency and precision, flexible in operation and the like.

Description

technical field [0001] The invention relates to the technical field of small target detection in small unmanned platforms, in particular to a deep learning small target detection method and device based on cascade fusion and attention mechanism. Background technique [0002] In standard detection datasets, large and medium objects usually occupy a large proportion. Some current detection algorithms have achieved a good level of detection effect on large and medium targets. However, the environment faced in practical application scenarios is often more complex. When the distance between the imaging device and the target is far away or the target itself is small in size, the number of small targets will increase sharply. Therefore, the research on small target detection is important in many practical problems. has great application potential. For example, in automatic driving, accurate detection of distant vehicles, signal lights, and signs in advance is beneficial to expand...

AI Technical Summary

Problems solved by technology

When using methods such as deep learning for detection, in order to reduce the amount of calculation and ensure the detection speed, the original image is usually further compressed, which will cause more information loss

How to effectively detect small targets is a big challenge

[0005] (2) Weak texture information

In the traditional deep learning method to achieve target detection, it usually includes four steps: image preprocessing, feature extraction, classification and regression, and post-processing. Feature extraction is implemented using a deep convolutional neural network, which uses shallow to deep levels Features, different levels of feature maps in deep convolutional neural networks correspond to different receptive fields, and the degree of information abstraction they express is also different. Large and medium-sized targets can easily obtain enough information from higher-level single-layer feature maps, but Small targets are difficult to obtain good detection results from the information of single-layer feature maps, so the shallow to deep hierarchical features in traditional deep convolutional neural networks are difficult to meet the needs of small target detection

[0007] To sum up, at present, the detection is usually carried out for large-sized or medium-sized targets, and there is no method that can accurately and quickly realize the detection of small-sized targets smaller than conventional sizes. Therefore, it is urgent to provide a deep learning small target detection method. Enables efficient small object detection based on deep learning methods

Images