Object recognition method and system, terminal equipment and storage medium

Abstract

The invention discloses an object recognition method and system, terminal equipment and a storage medium, and the object recognition method comprises the steps: S1, collecting picture data sets of different objects in an indoor real environment, calibrating pictures, and dividing the pictures into a training set and a test set; S2, replacing a part of convolutional layers in the backbone network of the YOLOv4_tiny network model with an inverse residual layer, and adding an SPP layer at the tail end of the backbone network of the YOLOv4_tiny network model to obtain an improved lightweight network model; S3, training and testing the improved lightweight network model by using the training set and the test set to obtain a trained lightweight network model; and S4, using the trained lightweight network model to identify an indoor object. According to the method, the target detection precision is remarkably improved on the premise that the model calculation amount and the parameter storage amount are hardly increased, and the detection performance of multi-scale targets, especially small targets, is improved.

Description

technical field [0001] The present invention relates to the technical field of computer vision, in particular to an object recognition method, system, terminal equipment and storage medium. Background technique [0002] Object detection is a basic task in the field of computer vision, which not only needs to classify the objects in the image, but also requires precise positioning. Object detection has a wide range of applications in many fields such as intelligent security, unmanned driving, home life, and service robots. With the continuous improvement of computer computing performance, object detection algorithms based on deep learning have gradually become mainstream. While the target detection performance has been improved, the shortcomings of the deep neural network itself, such as large parameter scale and high computational complexity, have become more and more obvious. In order to promote the deployment of deep learning networks on the embedded side, lightweight net...

AI Technical Summary

Problems solved by technology

(2) The other is to use lightweight thinking in model design, and try to improve the efficiency of the network in a limited number of layers. Although the number of parameters and calculations are greatly reduced, it is premised on sacrificing a certain detection accuracy.

[0011] YOLOv4-tiny has few parameters, although the detection speed is very high, but at the expense of accuracy

First, the lightweight network has fewer layers and cannot extract deep semantic information, which is extremely important for the positioning involved in target detection; second, when the input size is 416×416, the network’s prediction output layer It is 13×13 and 26×26, which reduces a very important prediction output layer 52×52 compared to YOLOv4, which leads to a decrease in the detection performance of multi-scale objects, especially the detection of small objects

Images