Edge computing device with edge learning image classification function and edge learning image classification method applied thereto

Abstract

This invention relates to an edge computing device with edge learning image classification capabilities and an edge learning image classification method applied thereto. The method is applied to a central server carrying a first neural network. The edge computing device carries a second neural network. The method includes the following steps: the central server performs a pre-training procedure using the first neural network to generate a first classification model set; the edge computing device connects to the central server and downloads the first classification model set; the edge computing device performs an edge learning procedure using the second neural network to generate a second classification model set; and the edge computing device performs an inference procedure on a target image based on the first and second classification model sets to generate an edge inference result.

Claims

1. An edge learning image classification method, applied to an edge computing device and a central server, the central server carrying a first neural network and the edge computing device carrying a second neural network, the method comprising the following steps: the central server performing a pre-training procedure using the first neural network to generate a first classification model set; the edge computing device connecting to the central server and downloading the first classification model set; the edge computing device performing an edge learning procedure using the second neural network to generate a second classification model set; and the edge computing device performing an inference procedure on a target image based on the first classification model set and the second classification model set to generate an edge inference result, wherein the method further comprises the following steps: the edge computing device inputting an incremental image data set into the second neural network for an incremental learning procedure; and the second neural network performing an incremental feature extraction procedure on the incremental image data set based on the first classification model set to generate an incremental feature extraction result; The edge computing device performs an incremental classification procedure on the first classification model set, the second classification model set, and the incremental feature extraction result to generate an incremental classification model set; and the edge computing device performs the inference procedure on the image to be detected based on the first classification model set, the second classification model set, and the incremental classification model set to generate another edge inference result.

2. The edge learning image classification method as described in Request 1, wherein the first classification model set has multiple classification models that classify image features based on shape, color, texture, location, object outline, or local structure.

3. The edge learning image classification method of request item 2, wherein the second classification model set has at least one classification model that classifies image features by shape, color, texture, position, object outline or local structure, and the number of classification models in the first classification model set is greater than the number of classification models in the second classification model set.

4. The edge learning image classification method as described in claim 1, wherein the image to be detected is acquired by the edge computing device by taking a picture of an object to be detected using a camera module, or is image data stored in a memory unit by the edge computing device.

5. The edge learning image classification method of claim 1, wherein the pre-training procedure includes the following steps: inputting a first image dataset into the first neural network by the central server; performing a first feature extraction procedure on the first image dataset by the first neural network to generate a first feature extraction result; and performing a first classification procedure on the first feature extraction result by the central server to generate the first classification model set.

6. The edge learning image classification method of claim 1, wherein the edge learning procedure includes the following steps: inputting a second set of image data into the second neural network by the edge computing device; performing a second feature extraction procedure on the second set of image data by the second neural network according to the first classification model set to generate a second feature extraction result; and performing a second classification procedure on the second feature extraction result by the edge computing device to generate the second classification model set.

7. The edge learning image classification method as described in requests 6 and 5, wherein the number of images in the second image dataset is less than the number of images in the first image dataset.

8. The edge learning image classification method as described in request item 1, wherein the computational level of the second neural network is lower than that of the first neural network.

9. The edge learning image classification method of claim 1, wherein the edge computing device is a user mobile device and the edge computing device is connected to the central server via an Internet connection.

10. The edge learning image classification method of claim 1, wherein the inference procedure includes the following steps: the second neural network performs a third feature extraction procedure on the image to be detected based on the first classification model set to generate a third feature extraction result; and the second neural network inputs the third feature extraction result into the second classification model set for inference to generate the edge inference result.

Images