Algorithm for counting articles by using multi-label network

Abstract

The invention discloses an algorithm for counting the number of articles by using a multi-label network. The algorithm comprises the following steps that: a client arranges the articles, wherein the client sends an arrangement request to a container; obtaining a tallying graph: the container receives the tallying request, the fisheye lens shoots a goods picture, and the goods picture is cut and repaired to obtain the tallying graph; multi-label network calculation: inputting the tallying graph into a mobile NetV2 network structure, adopting MSE as a loss function, and outputting the number ofarticles on each layer in the container after calculation; returning the number of commodities; the output data is returned to the corresponding commodity database; complete tallying, the method is reasonable in design, adopts the mobile NetV2 multi-label classification network and the non-object detection network, does not depend on detection box statistical quantity information, utilizes a neural network learning algorithm to obtain SKU quantity information, avoids statistical errors, and can save GPU resources and operation time.

Description

technical field [0001] The invention relates to the technical field of computer vision, in particular to an algorithm for counting the number of items using a multi-label network. Background technique [0002] Merchants tally goods and generate tally lists. Quantity statistics for each commodity are required to facilitate merchants to record commodity sales and check sales. [0003] The prior art uses the object detection network for item detection and quantity statistics. Object detection refers to the use of theories and methods in the fields of image processing and pattern recognition to detect target objects existing in images, determine the semantic categories of these target objects, and calibrate them. the position of the target object in the image. Object detection is a prerequisite for object recognition. Objects can only be recognized when an object is detected. However, when the object detection network performs object detection, it needs to rely on the statisti...

AI Technical Summary

Problems solved by technology

Only when the object is detected can the object be recognized, but the object detection network needs to rely on the statistical quantity information of the detection frame when performing object detection, which may cause statistical errors, and the image distortion of the fisheye lens currently used in the smart container project is relatively large , the detection effect is not good. For the tally list, the number of SKUs in each tally list is still counted manually. The speed of manually counting the number of SKUs is slow, and there is a possibility of missing records, which has certain defects.

Images