Risk identification and protection of electrified operation is of great significance to ensure the safety of personnel and the safe operation of power grid. Existing researches on risk identification mainly focus on the standardized wearing of safety protective tools

identification of unsafe dynamic behaviors of personnel

and detection of potential faults in electrical equipment

all of which do not sufficiently consider the interaction relationships between scene elements but address the risks in single scene elements. To this end

the paper proposes an interaction relationship analysis method for energized work scenario elements based on the prediction of human object interactions (HOI) sets. The method firstly perceives the downscaled flattened features of the electrified operation image based on the visual feature extraction module of the residual network

and obtains the position information of the image pixel features by using the position encoding module. Then

based on the multi-head attention mechanism encoder

enhanced feature vectors capturing global interaction relationships are obtained

and the decoder is used to generate a fixed number of HOI interaction feature query vectors. The iterative decoding of the query vectors is achieved through cross-perception of the enhanced feature vectors. Subsequently

an optimal interaction relationship is determined by using a multi-layer perceptron ensemble prediction module based on a bipartite graph matching approach

thereby achieving a detailed analysis of the interaction relationships of elements in live-line operation scenarios. Finally

the high-altitude operation scenario is taken as an example

and the results show that the proposed method achieves an average accuracy of 94.40% and demonstrates good recognition performance in various environments. The proposed model can serve as a reference for detecting interaction relationships in live-line scenarios.