Domestic garbage image recognition method based on knowledge distillation learning

Abstract

The invention discloses a household garbage image recognition method based on knowledge distillation learning, and relates to the technical field of garbage classification and data information processing. The method comprises the following steps: setting different acquisition variables for simulating the surrounding environment of garbage in a real state; through fine adjustment of the teacher model, the teacher model and the student model have the same network detection head, so that the teacher model and the student model can perform distillation learning. Training the fine-tuned teacher model and student model on a data set to obtain weight parameters; respectively designing loss functions for the backbone network and the network detection head, loading obtained weight parameters, and carrying out distillation learning; and finally outputting a recognition result. According to the invention, the recognition precision of the small parameter quantity model on the garbage image is improved through the construction of the household garbage image data set and the knowledge distillation method, so that the rapid and accurate recognition of the household garbage image is realized.

Description

technical field [0001] The invention relates to the technical field of garbage classification and data information processing, in particular to a method for recognizing images of domestic garbage based on knowledge distillation learning. Background technique [0002] In recent years, the revival of artificial intelligence technology and the continuous development of related theories have provided new opportunities for other disciplines to overcome difficulties, such as medical care, finance, education and other fields, and play an increasingly important role in these fields, has an increasingly wide range of applications. Deep learning in artificial intelligence can learn the inherent laws and representation levels of sample data, and can recognize data such as text, images, and sounds. The advantages of deep learning in image recognition enable garbage image recognition based on deep learning to solve the problem of household garbage classification. [0003] Although hous...

AI Technical Summary

Problems solved by technology

Taking Faster R-CNN as an example, as a second-order target detection algorithm, although it has good accuracy, the model faces the shortcomings of large parameters and slow inference speed.

Today, as smart garbage classification becomes more and more popular, deploying a model with a large number of parameters to the mobile terminal requires high cost, while small models face the problem of insufficient accuracy.

In addition, the labeled and publicly available datasets required for deep learning are currently unclear, which severely limits the development of research areas such as object recognition and garbage classification.

[0005] At the same time, the learning of the deep learning system requires a large amount of data in order to have better performance, and the garbage image datasets available for research are currently scarce.

On the other hand, although some deep models maintain high accuracy, the number of parameters of the model is extremely large, which brings great difficulty to the deployment of the model on the mobile terminal.

Images