Deep learning training recognition method

Abstract

The invention discloses a deep learning training recognition method, and the method comprises the following steps: carrying out the first target positioning training of a sample image through employing a deep learning model, extracting a target from the sample image, calibrating the target, and generating a target positioning model through training; classifying the samples to generate a classifiedsample set; identifying targets in the classification sample set by using the target positioning model trained for the first time, taking out an identified target area image, and storing the identified target area image as a new classification set sample; performing second training on the new classification set samples and the classification sample set by adopting a network model for second training to generate a classification recognition model. In the recognition process, a target is recognized by the target positioning model, and then classification recognition is performed through the classification recognition model. According to the training recognition method for deep learning, the recognition method of first positioning and then classification is adopted, and the recognition rateof deep learning can be effectively improved.

Description

Technical field [0001] The present invention relates to the field of artificial intelligence technology, in particular to a training recognition method for deep learning. Background technique [0002] Deep learning is a branch of artificial intelligence. Many fields are now being researched and applied. It is a process of prediction and recognition. Deep learning is first divided into two stages: training and prediction. Training uses machine learning to automatically identify features to complete the feature extraction of the target. Prediction is to identify the target in the sample through the trained model. [0003] The recognition rate of the deep learning generative model is close to that of humans, which is mainly due to the development of the following three aspects: 1. Algorithm improvement and improvement; 2. Big data support; 3. Computer operation efficiency improvement. However, the recognition rate of deep learning is still low. [0004] Target positioning is developed...

AI Technical Summary

Problems solved by technology

However, the recognition rate of deep learning is still low

[0004] Target positioning is developed based on target classification. The usual practice is to integrate classification and positioning in target positioning. The problem is that target positioning depends on limiting factors such as target classification and target size, which leads to a decline in recognition rate.

Images