A method of image target detection based on dc-spp-yolo

Abstract

The invention discloses an image target detection method based on DC-SPP-YOLO. Firstly, a data enhancement method is used to preprocess the training image samples and a training sample set is constructed, and a k-means clustering algorithm is used to select a target bounding box prediction The prior candidate frame of the YOLOv2 model; then the convolutional layer connection method of the YOLOv2 model is improved from layer-by-layer connection to dense connection, and at the same time, spatial pyramid pooling is introduced between the convolution module and the target detection layer to establish DC-SPP-YOLO target detection Model; finally, the loss function is constructed by the sum of squared errors between the predicted value and the real value, and the weight parameters of the model are updated iteratively to make the loss function converge, and the DC‑SPP‑YOLO model is obtained and used for target detection. The present invention considers the "gradient disappearance" caused by deepening the convolutional network and the YOLOv2 model does not fully use the multi-scale local area features, and constructs the improved DC-SPP-YOLO target detection model based on the dense connection of the convolutional layer and the spatial pyramid pooling. Improved object detection accuracy.

Description

technical field [0001] The present invention relates to an image target detection method, which belongs to the technical field of machine vision, and in particular to a target detection method based on dense connection and spatial pyramid pooling YOLO (Dense Connectivity and Spatial Pyramid Pooling Improved You Look Only Once, DC-SPP-YOLO) . Background technique [0002] Object detection is one of the core research contents in the field of machine vision, and it is widely used in driving navigation, workpiece detection, robotic arm grasping, etc. Establish and train a high-quality target detection model, which can extract more abundant and effective target features, and improve the accuracy of locating and classifying targets in images or videos. [0003] Traditional target detection methods such as Deformable Parts Models (DPM) search for target positions through sliding windows, which is inefficient; extracting artificially designed features such as histogram of oriented ...

AI Technical Summary

Problems solved by technology

However, the YOLO and YOLOv2 methods still have limitations in the target detection accuracy; when the model learning ability is improved by deepening the network, the gradient disappearance phenomenon will occur. The YOLOv3 algorithm uses the residual connection method to alleviate the gradient disappearance phenomenon but hinders the information of each layer of the network. At the same time, the multi-scale target detection of YOLOv2 and YOLOv3 algorithms focuses on the fusion of global features of different scale convolutional layers, ignoring the fusion of local area features of different scales in the same convolutional layer; this restricts the improvement of target detection accuracy

Images