Research On Weather Fax Image Recognition Technology

The maritime climate is changeable.Ships need to obtain first-hand weather information when sailing,and make corresponding route planning and emergency plans based on real-time weather conditions.The meteorological fax map is a real-time meteorological information carrier transmitted through remote meteorological stations.During the voyage,shipboard personnel can obtain the real-time weather conditions by analyzing the meteorological fax images,which plays a vital role in guiding navigation decisions.Using modern digital image processing technology and deep learning technology to complete the analysis of meteorological facsimile maps can improve the efficiency and accuracy of image recognition for shipboard personnel at sea.This subject takes meteorological fax images as the research object,analyzes the characteristics of the images and the actual work requirements,studies and analyzes a series of image processing methods such as image thinning,denoising,and scaling.At the same time,it integrates the results of the study and proposes a latitude and longitude network and equivalent Line extraction algorithm,and using modern deep learning algorithms to achieve image feature recognition and target detection.A complete weather fax image processing and identification system has been formed,which has certain reference function and application value for the research and application of weather fax images.First of all,according to the needs of image recognition,the pre-processing tasks such as image thinning,image noise reduction and image scaling were completed.In terms of image refinement,combined with the Zhang-Suen refinement algorithm,a refinement algorithm suitable for meteorological fax images is implemented.In terms of image noise reduction,the scale characteristics and distribution characteristics of image noise are analyzed.First,the traditional median filtering algorithm is adopted.Then,based on the principles of traditional methods,a threshold filtering algorithm and a connected domain count filtering algorithm are proposed.The counting method has the best noise reduction effect for weather fax images and can be used as a sample processing preprocessing algorithm.In terms of image scaling,the nearest neighbor interpolation algorithm,bilinear interpolation algorithm,and bicubic convolution algorithm were used for experiments.After comparing the results,the bicubic convolution algorithm has the best visual effect,but the calculation speed is slower than the previous two.Subsequently,the image processing algorithm was integrated to complete the contour,warp and weft network extraction and numerical recognition.Based on the Hough transform,a straight line extraction algorithm is implemented to complete the extraction of the latitude and longitude grid,and the voting threshold is appropriately raised to reduce the misjudgment of straight lines.Based on the idea of depthfirst search in algorithmic graph theory,an eight-neighbor depth-first search algorithm was proposed to identify and distinguish different contours.For the problem of the intersection of the latitude and longitude grid and the contour line,a double standard fuse method was proposed to reduce the background interference of the latitude and longitude network..Analyze the numerical distribution position and numerical morphological standard,complete the image segmentation,and identify the image numerical value based on the template matching algorithm.For clear morphological images,contour extraction and numerical extraction can be achieved.Finally,for meteorological fax images containing meteorological features such as highpressure centers,low-pressure centers,and ocean fronts,use the deep learning target detection framework YOLOV3 proposed in the second half of 2018 to perform target detection experiments,modify configuration parameters,and explore the parameters applicable to this subject,Verified by experiments,YOLOV3YOLO framework is suitable for small object detection,and has good adaptability for meteorological fax target detection.Based on the YOLOV3 algorithm framework,this topic builds a deep learning model,analyzes and trains the Loss curve,PR curve,and m AP value.The model has good evaluation indicators and high recognition accuracy.Finally,the sample is expanded,modified parameters,and K-Means clustering method is used to conduct experiments Improve.In the field of meteorological fax image recognition,the combination with deep learning methods was completed for the first time,which has high application value and pioneering significance.