Smoke video detection method and system based on improved MobileNetV2-SSD

Abstract

The invention discloses a smoke video detection method based on an improved MobileNetV2-SSD, and the method comprises the steps: (1) carrying out the optimization on an SSD target detection framework,and obtaining a MobileNetV2-SSD model; (2) on a MobileNetV2-SSD target detection framework, carrying out pyramid feature map reconstruction operation on the MobileNetV2-SSD target detection framework; (3) setting default candidate box parameters of each feature layer of the model according to the priori features of the smoke; (4) embedding an SE-Net module behind the six feature layers, and an improved MobiletV2-SSD model is obtained; and (5) training an improved MobileNetV2-SSD model by using the smoke data set to obtain a smoke detection model (SDM), which is used for detecting a smoke video. The invention further discloses a system based on the method. The invention provides a more effective method for smoke detection, can be widely applied to terminal equipment of an intelligent system without depending on a network environment, and has a very wide application scene.

Description

technical field [0001] The invention relates to the field of image processing, in particular to a smoke video detection method and system based on the improved MobileNetV2-SSD. Background technique [0002] The smoke video detection method refers to identifying and locating the smoke that appears in the video with the help of intelligent algorithms, so as to determine whether smoke occurs in the video frame. The video-based smoke detection method can obtain rich video information, so that it can detect smoke more intuitively and quickly. It is not limited by the size of the environmental space and has stronger adaptability, which is conducive to timely and effective fire control. At present, the method based on deep learning is mainly used for smoke detection, and the characteristics of smoke are learned independently through the neural network, which can effectively improve the accuracy of detection. [0003] However, due to the diverse characteristics of the shape, color,...

AI Technical Summary

Problems solved by technology

[0003] However, due to the diverse characteristics of the shape, color, and movement of smoke, that is, there is no fixed pattern, which increases the difficulty of detection

At the same time, due to the unpredictability of the environment and the complexity of the background when the smoke occurs, the smoke data set is still incomplete, which brings certain difficulties to the smoke video detection technology based on deep learning.

[0004] At present, there are relatively few video surveillance systems or intelligent terminal devices with smoke detection functions on the market. Most smoke detection systems based on deep learning are still in the research stage and have not been put into use on the market. The main reason is: In order to obtain better detection As a result, the number of layers in the network is increasing, and the parameters are also tens of thousands, resulting in the detection relying more on the cloud server and relying too much on the network, which is difficult to apply in real life

Embedded in the mobile terminal, it is even more difficult to realize the production of smoke detection products

Images