Unmanned aerial vehicle image real-time target detection method
A technology for target detection and drones, applied in the fields of image processing, computer vision, and pattern recognition, can solve problems such as large scale changes, unbalanced aspect ratios, narrow lengths, etc., achieve good detection accuracy and efficiency, and achieve detection accuracy and speed effect
- Summary
- Abstract
- Description
- Claims
- Application Information
AI Technical Summary
Problems solved by technology
Method used
Image
Examples
Embodiment 1
[0056] Embodiment 1: network composition structure
[0057] see figure 1 , the target detection network structure based on strip pooling is divided into three sub-structures, namely backbone network, feature extraction network and detection head. It should be noted that the basic components of the backbone network and the feature extraction network are constructed by the improved pooling bottleneck layer. These three parts are described below.
[0058] 1. By figure 1 As shown in the left half, the backbone network follows YOLOV5's backbone network DarknetCSP, and P1-P5 are different feature layers of the backbone network. The feature layer is mainly composed of BottleneckCSP (cross-domain connection bottleneck layer) stacked. This structure consists of the classic residual structure - showing a 1x1 convolutional layer (convolution block + batch normalization + activation function), then a 3x3 convolutional layer, and finally through the residual structure with The initial...
Embodiment 2
[0064] Embodiment 2: the unmanned aerial vehicle image real-time target detection method based on strip pooling, comprises the following steps:
[0065] 1. Perform data enhancement operations on UAV images, including rotation, translation, scaling, and blending. For training, validation, and testing of object detection networks.
[0066] 2. Network model design, modification and construction, including the following three items:
[0067] 1. According to the characteristics of large scale changes, unbalanced aspect ratio, and dense distribution of the target to be detected in the UAV image, the basic network structure of YOLOV5 is improved and designed. details as follows:
[0068] The strip pooling operation is introduced to better extract the spatial information of objects in the image and obtain long-distance dependencies in one direction, so as to improve the detection accuracy of objects with unbalanced aspect ratios. Under the guidance of the idea of the strip poolin...
PUM
Login to View More Abstract
Description
Claims
Application Information
Login to View More