Tire image recognition method and tire image recognition device

1. A tire image recognition method, comprising:

a step of obtaining a plurality of images of tires that differ from one another in either one of or both of a tire type and a tire condition, the obtained images being regarded as teacher images;

a step of cutting out, from the teacher images, a plurality of images having a size smaller than that of the teacher images and having a size of a predetermined number of pixels, and converting the cut-out images so as to obtain converted images of learning data;

a step of learning by a convolutional neural network using the converted images of learning data as learning images, and setting parameters for the convolutional neural network;

a step of obtaining a tire image of a recognition-target tire, cutting out, from the obtained tire image, a plurality of images having a size identical to that of the converted images of learning data, and converting the cut-out images so as to obtain converted images of recognition data; and

a step of inputting the plurality of converted images of recognition data to the convolutional neural network and determining, for each of the plurality of inputted converted images of recognition data, either one of or both of the tire type and the tire condition of the recognition-target tire.

2. The tire image recognition method according to claim 1 , wherein, the tire type or the tire condition is either one of a tread pattern, a tread wear amount and a clack in a side tread.

3. The tire image recognition method according to claim 1 , wherein at least one pattern periodic structure is captured in the teacher images and the tire mage of the recognition-target tire.

4. The tire image recognition method according to claim 1 , wherein, the converted images of learning data and the converted images of recognition data are converted into gray scale, and gradations of the gray scale are normalized in a range of 0 to 1.

5. A tire image recognition device, comprising:

a tire image capturing means that captures a plurality of teacher images and a recognition-target image, the teacher images being images of tires that differ from one another in either one of or both of a tire type and a tire condition; and

a computer including a storage, the computer configured to cut out, from the teacher images, a plurality of images having a size smaller than that of the teacher images and having a size of a predetermined number of pixels; convert the cut-out images so as to obtain converted images of learning data; cut out, from the recognition-target image, a plurality of images having a size identical to that of the converted images of learning data; and convert the cut-out images so as to obtain converted images of recognition data;

extract feature amounts of the converted images of learning data and the converted images of recognition data; and

compare feature amounts of the converted images of learning data with feature amounts the converted images of recognition data to determine either one of or the both of the tire type and the tire condition of the target tire,

wherein the feature amount extracting includes a convolution layer and a pooling layer of a convolutional neural network in which the converted images of learning data are configured as learning images, and

wherein the storage includes a fully connected layer of the convolutional neural network, the fully connected layer of the convolutional neural network compares the feature amounts of the converted images of the learning data with the feature amounts of the converted images of recognition data.

6. The tire image recognition method according to claim 2 , wherein, the converted images of learning data and the converted images of recognition data are converted into gray scale, and gradations of the gray scale are normalized in a range of 0 to 1.