Vehicle Information Detection Based On Deep Learning

Identification of car models,logos,license plates,car colors,etc.through artificial intelligence algorithms is often important in automotive-related industries in the era of big data.Deep learning algorithms are currently widely used for their outstanding performance in visual recognition.High-quality features can be extracted through deep convolutional neural networks,such as InceptionV2,ResNet,etc.The Inception network focuses on designing different functional convolution algorithms in the convolutional layer,to enhance its expression ability.and the improved version reduces the amount of parameters to facilitate training.The ResNet network focuses on increasing depth and constantly abstracting features to enhance network performance.According to the feature,we use the algorithm to complete the object localization and recognition,which is called the object detection algorithm,such as Faster-RCNN,SSD,etc.Faster-RCNN has high accuracy but slow speed because it has two detections,the first coarse detection and the second careful detection.SSD detects one-stage detection on multiple convolution features,ensuring accuracy while achieving real-time detection.After completing the positioning of the vehicle and license plate,we count color in the car area to determine the car color,we make license plate correction and number segmentation to identify license plate numbers.We compared five algorithms FasterRCNN-InceptionV2,FasterRCNNResNet50,FasterRCNN-ResNet101,FasterRCNN-ResNet152,SSD-InceptionV2 through experiments,and found that the FasterRCNN-ResNet101 algorithm achieved the best detection result on the test set,with a map of 82.47%.