Instrument detection method based on one-shot mechanism

Abstract

The invention provides an instrument detection method based on a one-shot mechanism. The method comprises the following steps: firstly, performing enhancement processing on a collected instrument image data set by utilizing size transformation and perspective transformation; constructing an instrument detection network model based on a one-shot mechanism, wherein the instrument detection network model comprises a feature extraction twin network module based on ResNet-18, an RPN module, a ROIAlign pooling layer and a feature fusion network module based on Grid Head exchange; training the network model by using the enhanced image data set to obtain a trained network; and finally, processing a to-be-detected instrument image by using the trained network to obtain a final detection result. The method has the capability of detecting the dial position of any quadrilateral instrument, and has relatively high instrument detection precision.

Description

technical field [0001] The invention belongs to the technical field of computer vision and target detection, and in particular relates to an instrument detection method based on a one-shot mechanism. Background technique [0002] Instrument detection and positioning technology, before the introduction of deep learning, mainly uses image matching to achieve accurate positioning of instruments using instrument templates. However, this method relies heavily on the accuracy of image matching. Based on SIFT (Scale-invariant feature transform, scale Traditional features such as variable feature transformation) are difficult to adapt to complex conditions, such as changes in illumination, changes in image clarity, and changes in the appearance of instruments from different manufacturers, which lead to positioning failures in some scene conditions. However, using detection methods based on HOG (Histogram of Oriented Gradient, histogram of oriented gradients) and SVM (Support Vector ...

AI Technical Summary

Problems solved by technology

[0003] There are two main difficulties in using computer vision technology for instrument detection and positioning: First, instrument detection tasks often require accurate acquisition of the dial position of the instrument, and correcting it into a rectangle to facilitate subsequent reading recognition, etc., especially those obtained under monitoring equipment The instrument picture is often tilted at an angle, and the rectangular dial of the instrument needs to be transformed after perspective transformation. Therefore, the algorithm must have the ability to predict the position of any quadrilateral, and the general target detection technology usually only needs to predict the minimum bounding box of the target. , for the detection and positioning task of the instrument panel, its positioning effect is difficult to meet the requirements of the follow-up tasks, and it is impossible to further correct the instrument panel to the instrument panel under the front view angle by only using the position information of the minimum bounding box of the instrument panel

In addition, even if the instance segmentation technology can be used to obtain the position mask of the instrument panel, it is faced with the problem of how to reduce the error introduced when using the position mask for perspective transformation; second, in different production environments, the appearance of instruments with the same function has different There is a certain similarity, but due to different instrument manufacturers, different factory batches or different versions and models, the appearance of the instrument will change to a certain extent. The current detection algorithm based on deep learning relies on the data distribution of the training set. When the actual test instrument data occurs Significant changes, that is, the instrument appearance of the training set is different from that of the test set, which will have a greater impact on the test effect

Images