Research On Pipeline Fire Image Recognition Method Based On Convolutional Neural Network

With the progress of society and the development of urban underground comprehensive pipe corridor,urban underground comprehensive pipe corridor integrates various engineering pipelines such as electric power,communication,gas,heat supply,water supply and drainage,and plays an important social role.It is of great significance to find and extinguish the fire before it causes great loss.The traditional fire alarm method based on temperature sensor or smoke sensor may cause false alarm and omission due to a large amount of interference from environmental factors.With automatic fire detection alarm method of digital image processing technology using video tests are the camera to monitor the scene,and then using the designed digital image processing algorithms of image processing,then came out of the suspected area feature extraction,finally adopts the method of pattern recognition to fire image recognition.Therefore,in contrast,the fire detection method using image processing technology can monitor and obtain more on-site information,thus making the system more agile in responding to fire and more suitable for early fire detection system.In this paper,the image recognition method of pipe corridor fire based on convolutional neural network is deeply studied.The main research results of this paper are as follows:A fire image brightness enhancement method is studied to solve the problem of insufficient image brightness in pipe corridor.At the present stage,the image quality of pipe corridor fire is mainly damaged due to uneven brightness distribution and discrete point interference.To solve above problems,this article multimodal fusion technology was applied to improve the quality of utility tunnel fire image,first of all,the original RGB channels into HSV and isolated three independent channels are tonal,saturation and brightness channel,then the luminance channel histogram equalization processing,can significantly enhance brightness.Then,on the premise of image brightness enhancement,the median filtering operation is carried out on the image to remove the discrete value interference in the fire image,so as to effectively improve the clarity of the V channel.Finally,the three separated channels are fused and converted to RGB channel to effectively improve the image quality.In the image processing of pipe corridor fire with obviously low brightness value and more discrete noise,the proposed multi-channel fusion algorithm can improve the image quality obviously.A k-nearest neighbor algorithm based on Particle Swarm Optimization(PSO)is proposed to solve the problem of slow clustering and heavy computation.At present,the main problem of k-nearest neighbor algorithm in processing pipe corridor fire images is that the amount of fire image data is large,which leads to the large amount of computation and long time consumption of k-nearest neighbor algorithm.Clustering algorithm is sensitive to the initial value,different initial value will lead to different results.To solve the above problems,this paper firstly applies PSO to the k-median clustering algorithm.The advantage of PSO is that it applies the speed-displacement model to the global search strategy,which greatly reduces the computational complexity.Then the clustering number is determined to make the algorithm reach the maximum fitness and effectively improve the computational efficiency of the algorithm.In the application of image region segmentation of pipe corridor fire,the proposed algorithm can also achieve a better region segmentation effect when the sample set is large.An adaptive pooling algorithm based on convolutional neural network is studied.Traditional pooling methods of convolutional neural network include maximum pooling and mean pooling.These methods do not take into account the interference of extreme values and may cause loss of original image information in some cases.To solve the above problems,this paper first inputted the segmented images of the fire area into the convolutional neural network,so as to avoid the network from learning a large amount of non-fire characteristic information and reduce the training time.Then based on the adaptive pooling method,a dynamic pooling method selection model is established,which improves the stability of pooling operation.In the application of image recognition of pipe corridor fire,the proposed algorithm has higher fire recognition accuracy than the neural network using traditional pooling method.A fire image recognition software based on PyQt is designed.The software system mainly realizes the image recognition of pipe corridor fire,and mainly designs three modules,which are the main interface,the system home page and the login interface.In the main interface,the corresponding result display interface is designed according to the algorithm proposed in this paper,and the complete process of pipe corridor fire image recognition is realized.In the login interface,My SQL is used to establish a connection with the background database to verify the correctness of the login user information,so as toensure the security of the software login and use.For the operation of image recognition of pipe corridor fire,the designed software is easy to operate.