The invention relates to the technical field of
visual inspection, provides an AI
visual inspection method for bagged objects, and aims to improve the identification efficiency and accuracy of equipment such as an intelligent scale on commodities. The method comprises four main steps of
image acquisition and conversion, color stripping, bag shape
processing and
image detection. First, an image of a bagged object is acquired and converted from RGB to HSV
color space, while an image containing only bagged material is acquired and the same conversion is performed. Secondly, stripping the color in the HSV space through subtraction, and converting the result back to the RGB space; in the bag shape
processing stage, discontinuous areas are removed through gray value
processing and morphological operation such as expansion and
carving, and the maximum continuous area of the object is reserved. And finally, performing
feature extraction and object recognition on the preprocessed image by using a
convolutional neural network (CNN) model. According to the invention, bagged and unbagged articles can be identified through one-time training, so that the
training time and cost are remarkably reduced.