Lightweight target detection method

Abstract

The invention discloses a lightweight target detection method. The lightweight target detection method comprises the steps of 1, performing data enhancement processing on a sample image; obtaining a prior bounding box size of the network model; step 2, constructing a target detection network model: the target detection network model is based on YOLOv4, a MobileNetv3 network is introduced to reconstruct a feature extraction network, standard convolution is replaced by deep separable convolution in PANet, and model parameter quantity and operand are reduced; after convolution operation is carried out on the feature layer with the same channel number, an improved CBAM attention mechanism is integrated, and the network detection performance is further improved; 3, training a target detection network model; and 4, detecting by using the target detection network model to obtain a detection result. The method has the characteristics of improving the target detection efficiency and reducing the network prediction time.

Description

technical field [0001] The invention relates to the field of machine vision, in particular to a lightweight target detection method. Background technique [0002] The three main applications of machine vision in the industrial field are vision measurement, vision guidance and vision inspection. At present, the one-stage algorithm is often used for target detection in the intelligent production process. Some researchers have proposed a deep learning model YOLOv4-FPM for bridge crack detection, and optimized the loss function and network to improve the FPS of the network. However, there is a problem that the large amount of parameters is difficult to deploy on the device; there are also researchers who have proposed a lightweight convolutional neural network YOLOv4-Lite for fruit detection, which greatly reduces the amount of network parameters after replacing the backbone feature network, but Due to the reduction of the number of parameters, the detection accuracy rate is no...

AI Technical Summary

Problems solved by technology

In short, the existing target detection methods still have problems such as low detection accuracy and increased detection time.

Images