Research On Infrared Liquid Level Detection Technology Of Storage Tank Based On Deep Learning

In the chemical,petroleum,aerospace,medical,beverage and other field.The real-time and accurate monitoring and detection of the liquid level in the closed container is an important prerequisite to ensure the safety of production and automatic management of equipment.Generally,it is necessary to select the appropriate measurement method according to the specific measurement environment and requirements.The liquid level detection in some special environments has special requirements for the measurement method and the instrument sensor used.Some existing detection methods are becoming mature in detection accuracy and reliability technology.However,due to the detection principle of the method itself,it is necessary to pay attention to the installation position and angle.Measurements under high temperature and pressure,corrosion,flammability and other conditions will not only affect the service life of the sensor,but also have higher requirements on the design of the sensor,which will increase the use cost and subsequent maintenance cost.Infrared target imaging detection technology can realize non-contact and non-immersion collection of the storage tank and liquid state,and will not destroy the structural integrity of the container.Therefore,the thesis proposes an infrared liquid level detection method for storage tanks based on deep learning,and obtains an infrared liquid level detection model through optimization training,which provides a research basis for liquid level detection in daily use environment and special environment.First,infrared liquid acquisition system is seted up,use infrared acquisition equipment to take multi-angle shots of liquid storage tank models with under various percentage capacities liquid level states such as 10%,20%,100%,and preprocessing the collected infrared images.Because of the occlusion,blur,inconspicuous features,and different position angles that may appear in the actual use scene.A variety of image processing methods have been analyzed and tested,which can be used to process infrared images collected in different real environments to make the target features more obvious and detectable.Then,some optimization methods of the target detection network framework and liquid level detection algorithm are discussed.The first part prepares the data set needed for deep neural network training.The image data enhancement method is analyzed,the Mosaic data enhancement processing is used at the input of the model,and the network parameter update method of small batch gradient descent is used.The second part discusses and analyzes the model’s backbone network,neck part,detection head and optimization methods used in post-processing in detail,including feature extraction,activation function,prevention of overfitting,feature fusion optimization method,and frame regression Loss calculation method,non-maximum suppression method analysis,etc.The third part is order to set model hyperparameters,edit training files to train and debug the model many times,analyze model performance and quantify,and save a final model with mean average precision reach 0.804 when the IOU was 0.5.The model shows good robustness and recognition effect.Finally,the research discusses the liquid level detection system of liquid storage tank of infrared image based on deep learning.The composition and function of each part are introduced respectively,and the model obtained by training is used to perform inference detection on the infrared liquid level image and surveillance video of the storage tank.It can be seen from the experimental results that the model can accurately detect the position of the liquid storage tank,and predict the liquid level state obtained by reasoning,the percentage content of the liquid in the storage tank and the confidence level.Therefore,the detection method studied in the thesis can be applied to the liquid level identification and real-time monitoring of storage tank equipment in the chemical,petroleum,aerospace,etc,and has promising application prospects in the future intelligent upgrade of production.