Ship target real-time detection method and terminal based on improved SSD model

Abstract

The invention provides a ship target real-time detection method based on an improved SSD model, and the method comprises the steps: carrying out the preprocessing of a monitoring video ship data set, obtaining a preprocessed ship image, and enabling a preprocessing result to meet the preset training requirements of a detection network; inputting the preprocessed ship image into an improved SSD network for feature extraction, prediction box generation, classification and regression operation in sequence; and sending a classification and regression prediction result to a non-maximum suppression module, and obtaining a final detection result. Compared with a traditional method, on the basis of an original SSD network, convolution operation, up-sampling operation and a feature fusion mechanism are adopted, and a feature extraction framework of multi-scale feature fusion is formed. According to the method, high-resolution information of the low-level feature layer and deep semantic feature information are effectively combined to realize joint decision, so that higher classification accuracy and a better position regression effect are obtained.

Description

technical field [0001] The invention relates to the technical field of ship target detection, in particular to a real-time ship target detection method and terminal based on an improved SSD model. Background technique [0002] With the rapid development of the marine industry, real-time detection of intelligent ships plays an important role in maritime traffic safety and port management. Current ship detection methods mainly focus on remote sensing images or radar images, but due to the timeline of image acquisition, these methods are difficult to meet the real-time requirements in practical applications. In recent years, the increasing maturity of maritime monitoring equipment has provided a large number of visible light ship images and videos, which has greatly promoted the real-time detection of maritime ship targets. [0003] The SSD network has good robustness and is a commonly used one-stage object detector in the industry. Compared with Faster R-CNN and YOLO series,...

AI Technical Summary

Problems solved by technology

Maritime ship detection is a problem in the field of target detection. However, the traditional SSD model is not very effective in the field of real-time ship detection without the improvement of the network structure.

First of all, SSD detects targets of different scales through convolutional layers of different depths, but the low-level feature layers contain less semantic information, resulting in poor detection of small targets.

Secondly, although SSD can detect targets of different scales, it does not combine contextual information to further improve detection accuracy.

Finally, the size and aspect ratio of the prior frame of the SSD model cannot be directly obtained, but need to be manually set, so the detection efficiency is low, and the detection effect needs to be further improved

[0004] At present, there are not many researches on real-time detection of ship targets in the existing technology. Taking Faster R-CNN as an example, the detection speed of this detection method is very slow and the number of model parameters is large, which cannot be done on the existing common CPU equipment in various application scenarios. to real-time object detection

Images