Counting method for positioning dense products based on deep learning direction rectangles

Abstract

The invention provides a counting method for positioning dense products based on a deep learning direction rectangle, and the method is characterized in that the method comprises: an image data preprocessing stage: obtaining and cutting a product image, marking the direction rectangle position and category information of the product, and storing the position and category information; a model training stage: training parameters including an iteration period, a training set and verification set proportion and a data enhancement mode are set; starting training, when a loss function curve does not descend any more in the training process and a verification set IoU is larger than 0.75, stop training manually, or stop training automatically when the period reaches the set number of times, obtaining a model and storing same, and wherein IoU is the ratio of an intersection set and a union set of a model prediction position and a marked position area; and a deployment stage: importing the model obtained in the model training stage into an equipment industrial personal computer for operation, and carrying out product counting statistics on the charging tray placed on the equipment according to the model. The method has an efficient statistical effect, and the counting result does not need manual intervention.

Description

technical field [0001] The invention is applied in the field of automatic visual inspection, and specifically relates to a counting method for locating dense products based on deep learning direction rectangles. Background technique [0002] Electronic processors need to maintain enough memory for various components required for manufacturing, but piles of resistors or transistors and many other small parts are not only difficult to store, but also a big problem for quantity control. Users need to be able to accurately Only by controlling the current inventory quantity can we effectively manage storage conditions and costs. In an automated production line, trays are usually used to store components. The operator arranges a large number of components on the trays by pasting them regularly and then carries out assembly work. In order to know the number and type of components in each tray , it is necessary to efficiently count the number and types of components on the tray. Th...

AI Technical Summary

Problems solved by technology

[0002] Electronic processors need to maintain enough memory for various components required for manufacturing, but piles of resistors or transistors and many other small parts are not only difficult to store, but also a big problem for quantity control. Users need to be able to accurately Only by controlling the current inventory quantity can we effectively manage storage conditions and costs

In an automated production line, trays are usually used to hold components. The operator arranges a large number of components on the trays in a way of pasting regularly and then carries out the flow operation. In order to know the number and type of components in each tray , it is necessary to efficiently count the number and types of components on the tray. The traditional manual counting method has the disadvantages of time-consuming, high cost and large error, and it is impossible to accurately judge the actual quantity of incoming products.

Images