Lightweight improved target detection method and detection system

Abstract

The invention belongs to the technical field of computers, and discloses a lightweight improved target detection method and system, and the method comprises the steps: replacing an original backbone network in an FSSD with an improved ShuffleNet v2 model; introducing a weighted bidirectional feature pyramid structure to replace an original three-layer feature fusion structure in an FSSD, carryingout top-down and bottom-up feature fusion on multiple layers of features, so that the fused features contain more semantic information, and target detection is performed. Compared with a basic model,the model provided by the invention has the advantages that the accuracy is improved, the detection speed is reduced, and the weight of the model is increased to a certain extent. Through the analysisand the performance of experimental data, the improvement of the accuracy rate can be seen to accord with the improvement provided by the invention, and meanwhile, the defects in the aspects of detection speed and model weight caused by the improvement of the invention are also within a predictable range.

Description

technical field [0001] The invention belongs to the technical field of computers, and in particular relates to a lightweight and improved object detection method and detection system. Background technique [0002] At present, computer vision is a classic research field dedicated to using computers to complete human vision tasks. Among them, target detection is a basic research direction in this field. The main problem to be solved in this direction is to design corresponding algorithms so that the computer can locate and recognize objects of interest in the picture. With the development of image acquisition equipment and the improvement of the computing power of the equipment, the application of target detection and the image styles targeted by the research are more diverse, the resolution of the image is larger, the scenes in the image are more complex, and the most important thing is the image The amount of data is larger than before. As a member of data science, target ...

AI Technical Summary

Problems solved by technology

[0010] However, these algorithms also have the following disadvantages: manual extraction of image features is required, and certain professional knowledge requirements are required for the person who operates the feature extraction, which is inconvenient to use; the versatility of these algorithms is limited due to the method of feature extraction; There is still a gap between performance and practical application

[0058] Based on this shortcoming, many target detection applications that require real-time feedback results, such as unmanned driving, are difficult to achieve

[0059] (2) Most of the mobile smart devices used in target detection are embedded devices. Compared with servers with GPUs, such devices are very limited in storage space and computing power, which means that the weight of the target detection model and calculation The complexity of the limited

This makes it difficult to achieve real-time docking feedback for unmanned driving technology and pedestrian body shape monitoring that requires real-time applications.

[0060] (3) When improving the target detection algorithm model before, the accuracy, speed and weight of the model cannot all achieve a reasonable balance

It is also impossible to achieve the optimal solution, making it difficult for various technologies that need to achieve real-time interaction to meet the needs of inventors

[0061] (4) In the prior art, MobileNet (mobile network) is used to replace the VGG-16 network in the FSSD model as the backbone network of the model. This method has achieved certain advantages in the speed of the model and the size of the model weight, but the model's Relatively lower detection accuracy

[0064] (2) But at the same time, if we solve the problem without taking up storage space, we will be forced to reduce the network proportion of the algorithm part, which will reduce the accuracy rate, which makes us have to use new means to solve the problem

Images