CN-115937668-B - Country geographic tag updating method based on social media picture data

CN115937668BCN 115937668 BCN115937668 BCN 115937668BCN-115937668-B

Abstract

Aiming at the limitations of the prior art, the invention provides a rural geographic tag updating method based on social media picture data, which combines a plurality of deep learning methods, improves the prediction accuracy on the premise of ensuring the prediction efficiency, can automatically update the picture tags, simultaneously classifies various different scenes, has wide applicability, effectively improves the rural geographic tag updating efficiency, and reduces the labor cost and the working strength.

Inventors

LI YING
LI HUAN
Jiang Qianteng
WANG YONGTIAN
CHEN LUAN
LI MINSHENG

Assignees

奥格科技股份有限公司
中山大学

Dates

Publication Date: 20260512
Application Date: 20220824

Claims (9)

1. A rural geographic tag updating method based on social media picture data is characterized by comprising the following steps: s1, acquiring images to be predicted, which are acquired by social media and contain rural geographic content; s2, inputting the image to be predicted into a target detection network trained by a preset target detection data set, detecting rural facilities contained in the image to be predicted, and obtaining a classification result of the image to be predicted about a rural facility scene; After detecting the rural facilities contained in the image to be predicted, performing post-processing on mutual exclusion conditions in a classification result in the following manner: S21, judging whether the category of the classification result contains an agricultural house or a traditional building, if not, judging whether the ratio of the area of a detection frame of the agricultural house or the traditional building to the area of the image to be predicted is less than 5%, if yes, discarding the detection frame and the corresponding category, otherwise, executing the step S22; S22, judging whether the classification result comprises an agricultural house and a village commission or comprises a traditional building and a village commission at the same time, if not, judging whether a detection frame of the village commission is positioned in the agricultural house or the traditional building, if so, discarding the agricultural house classification or the traditional building classification and corresponding position information contained in the classification result, otherwise, not processing; S3, inputting the image to be predicted into a multi-label classification network trained by a preset multi-label classification data set to obtain a classification result of the image to be predicted about a rural environment scene; And S4, merging the classification results of the step S2 and the step S3, and finishing updating the country geographic tag of the image to be predicted.
2. The method for updating a rural geotag based on social media picture data according to claim 1, wherein the image data of the target detection dataset is marked with a data tag of a rural facility such as a traditional building, a rural house, a garbage collection point, a school, a public toilet, a village commission, and a village.
3. The method for updating a rural geographic tag based on social media picture data according to claim 2, wherein in the target detection network, the threshold of the category of villages is 0.2, the threshold of the category of garbage collection points is 0.35, and the thresholds of the categories of traditional buildings, rural houses, schools, public toilets and village commissions are all 0.25.
4. The method for updating a rural geographic tag based on social media picture data according to claim 1, wherein the image data of the multi-tag classification dataset is marked with data tags of rural environments such as roads, farmlands, water bodies and squares.
5. The method for updating a rural geotag based on social media picture data according to claim 4, wherein the multi-tag classification dataset is pre-processed prior to use in training by: Enhancing the image data of the multi-label classification dataset by adding gaussian noise, increasing brightness, decreasing brightness, increasing contrast, decreasing contrast; Normalizing the data characteristics of the image data of the multi-label classified data set to enable the value of the image matrix to be between 0 and 1; Data tags of image data of the multi-tag classification dataset are converted into an n-hot encoding format (1 < = n < = 4).
6. The method for updating a country geographic tag based on social media picture data according to claim 1, wherein the object detection network is a YOLO-v5 object detection network; The multi-label classification network is ResNet-50 multi-label classification network, the last layer of the multi-label classification network uses a sigmoid activation function, and an optimizer of the multi-label classification network is an Adam optimizer.
7. The rural geographic tag updating system based on social media picture data is characterized by comprising an image acquisition module (1) to be predicted, a rural facility scene classification module (2), a rural environment scene classification module (3) and a classification result merging module (4), wherein the image acquisition module (1) to be predicted is respectively connected with the rural facility scene classification module (2) and the rural environment scene classification module (3), and the classification result merging module (4) is respectively connected with the rural facility scene classification module (2) and the rural environment scene classification module (3), wherein: The image to be predicted acquisition module (1) is used for acquiring images to be predicted, which are acquired by social media and contain rural geographic content; The rural facility scene classification module (2) is used for inputting the image to be predicted into a target detection network trained by a preset target detection data set, detecting rural facilities contained in the image to be predicted, and obtaining a classification result of the image to be predicted on a rural facility scene; After detecting the rural facilities contained in the image to be predicted, performing post-processing on mutual exclusion conditions in a classification result in the following manner: S21, judging whether the category of the classification result contains an agricultural house or a traditional building, if not, judging whether the ratio of the area of a detection frame of the agricultural house or the traditional building to the area of the image to be predicted is less than 5%, if yes, discarding the detection frame and the corresponding category, otherwise, executing the step S22; S22, judging whether the classification result comprises an agricultural house and a village commission or comprises a traditional building and a village commission at the same time, if not, judging whether a detection frame of the village commission is positioned in the agricultural house or the traditional building, if so, discarding the agricultural house classification or the traditional building classification and corresponding position information contained in the classification result, otherwise, not processing; The rural environment scene classification module (3) is used for inputting the image to be predicted into a multi-label classification network trained by a preset multi-label classification data set to obtain a classification result of the image to be predicted on a rural environment scene; And the classification result merging module (4) is used for merging classification results of the rural facility scene classification module (2) and the rural environment scene classification module (3) to finish updating the rural geographic tag of the image to be predicted.
8. A storage medium having stored thereon a computer program, wherein the computer program when executed by a processor performs the steps of the social media picture data based country geotag updating method of any of claims 1 to 6.
9. A computer device comprising a storage medium, a processor and a computer program stored in the storage medium and executable by the processor, which when executed by the processor performs the steps of the social media picture data based country geotag updating method as claimed in any one of claims 1 to 6.

Description

Country geographic tag updating method based on social media picture data Technical Field The invention relates to the technical field of rural picture automatic processing, in particular to a rural geographic tag updating method based on social media picture data. Background With the rapid progress of internet technology and popularization of rural smartphones, a large amount of social media software goes deep into village people's daily life. The village scenery shooting applet has a plurality of functions of image-text social contact, questionnaire design, online scoring, online interpretation and the like, and a large amount of user and village data are accumulated in a whole country since the village scenery shooting applet is online. The image-text data has longitude and latitude information, can position the information to a specific position, and provides a novel observation means for the village living environment. The village scenery shooting provides a large number of village pictures with positioning, and can become an effective data source of village geographic POIs. The Chinese patent of the invention with the publication date 2021.11.02 discloses a method for classifying urban landscape elements based on street view pictures and measuring locality, which comprises the steps of firstly classifying the scenes of 365 subclasses of street view picture data of the cities, constructing a classification system of the locality landscape elements to achieve reclassification of the scenes, secondly measuring the locality level of each city from two angles of locality values and similarity values according to the classification system, and finally carrying out coupling analysis on the classified landscape elements and space elements such as road accessibility, facility density, functional diversity and the like, and describing the change process and action mechanism of the locality of the cities to judge the types of the landscape elements affecting locality differences and locality similarity. However, the data are mainly distributed in urban areas, and the geographic tag classification system is mainly concentrated in urban ground object scenes. Thus, the prior art has certain limitations. Disclosure of Invention Aiming at the limitation of the prior art, the invention provides a rural geographic tag updating method based on social media picture data, which adopts the following technical scheme: a rural geographic tag updating method based on social media picture data comprises the following steps: s1, acquiring images to be predicted, which are acquired by social media and contain rural geographic content; s2, inputting the image to be predicted into a target detection network trained by a preset target detection data set, detecting rural facilities contained in the image to be predicted, and obtaining a classification result of the image to be predicted about a rural facility scene; S3, inputting the image to be predicted into a multi-label classification network trained by a preset multi-label classification data set to obtain a classification result of the image to be predicted about a rural environment scene; And S4, merging the classification results of the step S2 and the step S3, and finishing updating the country geographic tag of the image to be predicted. Compared with the prior art, the rural geographic tag updating method based on social media picture data has the advantages that the prediction accuracy is improved on the premise of guaranteeing the prediction efficiency by combining multiple deep learning methods, the picture tags can be automatically updated, meanwhile, classification is carried out on multiple different scenes, the applicability is wide, the rural geographic tag updating efficiency is effectively improved, and the labor cost and the working strength are reduced. As a preferable scheme, the image data of the target detection data set is marked with the data labels of the rural facilities such as traditional buildings, rural houses, garbage collection points, schools, public toilets, village commissions and villages; further, in the target detection network, the threshold of the category of the village is 0.2, the threshold of the category of the garbage collection point is 0.35, and the thresholds of the categories of the traditional buildings, the agricultural houses, the schools, the public toilets and the village commissions are all 0.25. Further, in the step S2, after detecting the country facilities included in the image to be predicted, post-processing is performed on the mutually exclusive situation in the classification result by: S21, judging whether the category of the classification result contains an agricultural house or a traditional building, if not, judging whether the ratio of the area of a detection frame of the agricultural house or the traditional building to the area of the image to be predicted is less than 5%, if yes, discarding the detection fram