Semantic segmentation-based limb protection detection system and method

The invention discloses an edge protection detection system based on semantic segmentation, and the system is characterized in that a picture set obtaining module carries out the frame extraction of construction site video stream data collected historically and construction site video stream data collected in real time, and obtains a historical picture set and a real-time picture set; the data preprocessing module preprocesses the historical picture set; the data set division module divides the preprocessed historical picture set into a training set and a test set; an information labeling module performs pixel-level labeling on constructors and fences in the training set and the test set, and a model training module inputs the training set with labeling information into a deep neural network model to obtain a trained deep neural network model; the reasoning module obtains a segmentation result of the real-time picture set; the fence crossing judgment module judges whether the constructors cross the fence or n