A model fusion method suitable for beverage bottle recycling machine delivered object classification and identification

Abstract

The invention discloses a model fusion method suitable for beverage bottle recycling machine delivered object classification and recognition. The model fusion method comprises the following steps that1, based on the inception-v3 model, by using the bottleneck method, the deep convolutional neural network models model_1 and model_2 are obtained by training different data quantities, and the 24-layer structure of yolov3-tiny is modified into a 32-layer structure, and by training the specified amount of data, a deep convolutional neural network model model_3 is obtained; Step 2, on the same testset, the correct data sets of model_1, model_2, and model_3 are obtained, the correct data sets of model_1 and model_2 are taken together, and then the correct data sets of model_3 are combined to obtain the final identification data set.

Description

technical field [0001] The invention belongs to the technical field of article recycling, relates to a beverage bottle recycling machine, and in particular to a model fusion method suitable for classification and identification of deliverables in a beverage bottle recycling machine. Background technique [0002] At present, the classification and identification methods for the deliveries of beverage bottle recycling machines mainly include the following categories. The first one is based on general-purpose digital image processing technology, by taking digital images of beverage bottle recycling machine deliveries, analyzing its outline features, light intensity features, color features, local area template features, and deliverables’ attached barcode features and other information. One or more combinations of these methods are used to obtain a decision-making scheme for classification and identification, so as to classify and identify the deliverables of the beverage bottle...

AI Technical Summary

Problems solved by technology

[0003] In order to solve the problems that the digital image processing method is not effective in classifying and identifying deliveries in a variable sampling environment, the traditional machine learning method combined with the feature extraction method is too dependent on hardware, and the generalization ability of a single model trained using deep learning is general, the present invention Provide a model fusion method suitable for the classification and recognition of beverage bottle recycling machine deliveries. On the basis of a certain amount of data, use the transfer learning method to train two inception-v3 deep convolutional neural network models and one based on the yolov3-tiny structure The improved yolov3-tiny32 structure deep convolutional network model, the above three models are fused through the model fusion method, and training with less data can significantly improve the recognition accuracy of beverage bottle recycling machine deliveries, achieving a better recognition effect

Images