Target detection method based on deep learning

Abstract

The invention belongs to the technical field of machine learning, and particularly relates to a target detection method based on deep learning, and the method comprises the following steps: data collection: collecting a COCO 2014 detection data set and a PASCAL VOC 2007 + 2012 data set EMG to complete the construction of an original data set, marking the category of the original data set, and completing the construction of a data set needed by model training; data preprocessing: preprocessing the data, dividing different types of original data pictures through different data segmentation methods, and ensuring a model training effect; constructing a model: constructing an identification and classification model by adopting a low-dimensional convolutional neural network, inputting training data, completing the construction of a parameter model, and storing the model: when a loss function of the model is not reduced any more, storing the model; and model evaluation: inputting the test data into the stored network model to complete model performance evaluation. The invention provides a new normal form (G-RCN) for separating classification and positioning tasks, optimizes the gap between the classification and positioning tasks, and improves the performance.

Description

technical field [0001] The invention belongs to the technical field of machine learning, and in particular relates to a target detection method based on deep learning. Background technique [0002] At present, the deep neural network has achieved superhuman performance in the 1000-category target recognition task, but the detection network has not achieved human-level performance in the 80-category target recognition task, which shows that there is a gap between the recognition and detection tasks. Object detection requires strong classification performance and the ability to pinpoint an object among an infinite number of candidate positions. [0003] Problems or defects in the existing technology: Currently, object detection models use shared feature maps to simultaneously complete classification and localization tasks, but only by comparing the performance of the original fast RCNN and the RCNN with partially separated feature maps, and the performance is poor. SUMMARY O...

AI Technical Summary

Problems solved by technology

[0003] Problems or defects in the existing technology: At present, the target detection model uses the shared feature map to complete the classification and positioning tasks at the same time, but it can only compare the performance of the original fast RCNN and the RCNN with partly separated feature maps, and the performance is poor

Images