SSD-based target detection improved algorithm

Abstract

The invention discloses an SSD-based target detection improved algorithm. The SSD-based target detection improved algorithm comprises the steps of 1, inputting an image; 2, preprocessing the image input in the step 1; 3, sending the preprocessed image into a deep residual error network (ResNet) to carry out a series of convolution processing; 4, performing feature extraction by using a convolutional network, and extracting key information of the input picture; 5, sending the features extracted in the step 4 into an SSD network for re-extraction, and positioning and classifying the pictures; and 6, finally outputting a target detection result of the image. Compared with the prior art, the algorithm can be well used for target detection, the calculation amount is reduced, hardware memory resource consumption is reduced, the network model performance and the detection accuracy are improved, the detection accuracy is improved by 3.6% compared with a classic SSD algorithm, and good robustness is achieved. The method has important significance and reference value for further application of the deep learning technology to target detection.

Description

technical field [0001] The invention relates to the field of target detection, in particular to an improved algorithm for target detection based on SSD. Background technique [0002] Object detection has become an important research direction and research hotspot in the field of computer vision, and can be applied to unmanned driving, robots, video surveillance, pedestrian detection, sea ship detection and other fields. Before the emergence of deep learning, the target detection method was mainly completed by establishing a certain mathematical model based on certain prior knowledge. The widely used methods are: Hough transform, frame difference method, background subtraction method, optical flow method, sliding method, etc. Window models, deformable part models, etc. In recent years, the emergence of deep learning technology has revolutionized the mode of target detection and improved the accuracy and robustness of target detection. [0003] Currently, object detection me...

AI Technical Summary

Problems solved by technology

The detection performance of SSD is relatively better, but there are still many shortcomings: the calculation of SSD network parameters is relatively large, the performance of the network model is not good, and the detection accuracy of target detection needs to be improved

Images