Human activity recognition fusion method and system for ecological conservation redline

A human activity recognition fusion system for ecological protection red line, comprising:

• an obtaining pre-processing module, configured to obtain a pre-stage remote sensing image and a post-stage remote sensing image of a target ecological protection red line region, and to perform a data pre-processing;
• an input module, configured to input the pre-processed pre-stage remote sensing image and the post-stage remote sensing image into a human activity recognition model after a pre-training, wherein the human activity recognition model uses a plurality of sets of data training, the plurality of sets of data comprises:
• pre-stage images and post-stage images of a red line region, and the post-stage images carry a human activity annotation;
• an identification module, configured to identify a human activity pattern of the target ecological protection red line region as a first detection result;
• an analyzing and processing module, configured to segment, calculate and analyze latest image data corresponding to the target ecological protection red line region based on a geographical country situation data to obtain a change pattern as a second detection result, comprising:
• performing a guidance segmentation on the latest image data corresponding to the target ecological protection red line region based on the geographical country situation data;
• calculating spectral and geometric characteristics of each segmented pattern, and obtaining the change pattern by clustering analysis and quartile spacing method as the second detection result;
• a fusing module, configured to fuse the first detection result and the second detection result to obtain a change detection pattern of the target ecological protection red line region;
• wherein, the described calculating spectral and geometric characteristics of each segmented pattern, and obtaining the change pattern by clustering analysis and quartile spacing method as the second detection result specifically comprises:
• selecting a K-means algorithm as a clustering algorithm,
• adopting a similarity analysis to analyze factors of several characteristic parameters, and selecting X most heterogeneous characteristics from a plurality of characteristics to characterize the change;
• calculating a median M of parameters of the most heterogeneous characteristics, finding a value Xa at 25% forward from the median M,
• finding a value Xb at 75% backward from the median, and a quadratic difference Mk=Xb−Xa;
• taking values less than M−3*Mk and greater than M+3*Mk as a deviating region to obtain a change pattern of full sample.