Visual model training method based on OpenArt mini

The invention discloses a visual model training method based on OpenArt mini. The visual model training method specifically comprises the following steps: step 1, acquiring batch sample images by adopting an OpenArt mini visual sensor; step 2, carrying out augmentation processing on the obtained batch sample images; step 3, carrying out image processing on the sample image data set; step 4, training a model, and training the MobilenetV2 network by using the image after image processing, so as to obtain a trained MobileNetV2 network model; the OpenArt mini visual sensor detects an article and returns coordinates of the article in an image, the article is corrected to a correct position through perspective transformation and inverse perspective transformation, and the mobilenetV2 network model adopts EIQ Tookit to perform data augmentation operation on an input picture, so that the generalization ability of the model can be improved; according to the mobilenetV2 network model, a depth separable convolution meth