Multi-graphic target detection method based on optimized deep learning

Abstract

The invention discloses a multi-graphic target detection method based on optimized deep learning, and the method comprises the following steps: employing a marking tool to calibrate a data set needing to be recognized, and obtaining a calibrated data set; establishing a multi-target detection model; the multi-target detection model comprises a selection module and four different calculation modules; the four different calculation modules are respectively used for predicting circular, elliptical, inclined rectangular and triangular targets; the selection module is used for selecting one of the four calculation modules according to actual conditions; training the multi-target detection model by using the calibration data set, and obtaining a trained multi-target detection model by calculating iou of different calculation modules and optimizing a loss function; and completing detection by using the trained multi-target detection model. The method has the beneficial effects that compared with a plurality of independent target detection algorithms, the detection speed effect is improved, the processing speed of a single picture is averagely shortened by 18.8%, and the method is more suitable for industrial and engineering production processes.

Description

technical field [0001] The invention relates to the field of image target detection, and more specifically, relates to a multi-graphic target detection method based on optimized deep learning. Background technique [0002] The current mainstream image detection mainly uses median filtering to denoise, finds the outline of the image through the characteristics of the image, and then filters through judgment to obtain high-quality recognition results, and the characteristics of different images are also different. [0003] For example, a circle can use Canny edge detection and Hough circle to get the circle in the picture; an ellipse can obtain high-quality ellipses by splicing arc support line segments, ellipse clustering, and candidate ellipse verification, or use random Hough transform to identify ellipses and rectangles and convex polygons can also be identified by conventional algorithms. [0004] Most common target recognition algorithms use rectangular boxes to predict...

AI Technical Summary

Problems solved by technology

[0005] However, for some specific graphics, the rectangular frame obtained by the target recognition algorithm of the rectangular frame cannot fit the object well, and if the graphic detection is used for target recognition, it may be interfered by the same shape and different categories.

Images