Online detection device based on YOLO v5 model
A detection device and model technology, applied in biological neural network models, character and pattern recognition, image analysis, etc., can solve problems such as impact on production efficiency, impact on product quality or corporate image, inability to achieve one-by-one inspection, etc., to achieve savings human effect
- Summary
- Abstract
- Description
- Claims
- Application Information
AI Technical Summary
Problems solved by technology
Method used
Image
Examples
Embodiment 1
[0043] This embodiment provides an online detection device based on the YOLO v5 model, including
[0044] The backbone network BackBone uses the focus structure for slicing operations, then splices in the channel dimension, and finally performs convolution operations;
[0045] Feature extraction and fusion structure for fusion of different feature scales and transfer shallow features to the top layer;
[0046] A forecasting agency for making relevant forecasts for exploiting previous characteristics;
[0047] Among them, the backbone network BackBone, feature extraction and fusion structure and prediction mechanism jointly constitute the YOLO v5 model;
[0048] The judger judges whether the product is qualified according to whether a group of circles in the result of target detection are concentric.
[0049] The biggest feature of YOLO v5 in this embodiment is that it is fast and accurate, which is very suitable for mobile terminals, and its model is small and fast. Table 1...
Embodiment 2
[0057] On the basis of Embodiment 1, this embodiment performs a preferred operation. In the preferred aspect, the backbone network BackBone in this embodiment includes a focus structure, and the key in the Focus structure in this embodiment is the slicing operation. like image 3 As shown, it is similar to the reverse operation version of sub-pixel convolution. In short, the data is divided into 4 parts, each data is equivalent to 2 times of downsampling, and then spliced in the channel dimension, and finally Convolution operation. Its biggest advantage is that it can perform down-sampling operations to minimize the loss of information.
[0058] When this embodiment is preferably implemented, YOLO v5s defaults to 3*640*640 input, copies four copies, and then cuts the four pictures into four 3*320*320 slices through the slicing operation, and then uses concat Connect these four slices from the depth, the output is 12*320*320, and then pass the convolutional layer with 32 conv...
Embodiment 3
[0070] At a specific implementation level, this embodiment provides a method for position detection using target detection, including three processes of data labeling preprocessing (preprocessing), target detection (processing), and a judger (postprocessing). Its simulation process is as follows Figure 9 shown.
[0071] The preprocessing process of this embodiment includes data labeling (using labelImg to anchor the frame), data preprocessing (converting the label .xml format to .txt format and randomly dividing the training set and the validation set). The overall flow chart is as follows Figure 10 shown.
[0072] In the target detection (in processing) process in this embodiment, the main process in the processing process is to use the YOLO v5s model to perform the target detection operation. The specific process is shown in the overall structure of YOLO v5s. Among them, the training results of the YOLO v5s model are as follows Figure 11 shown. The visual image after...
PUM
Login to View More Abstract
Description
Claims
Application Information
Login to View More