Research On Depression Detection Based On Gait Skeleton Information

Depression is a very common mood disorder disease in the current society,which brings a significant burden to the society and individuals.However,the traditional diagnostic methods cannot achieve efficient diagnosis and screening of depression with large samples,making it an urgent need for a depression detection method with rapid,non-disturbed,objective and effective.In recent years,machine learning has achieved excellent research results in the detection of depression.Gait,as an emerging research method,has shown certain advantages in this research.Based on the above situation,this paper investigated depression detection using gait skeleton information.First,to better capture temporal and spatial information in gait movements,this paper presents a between-class learning probabilistic inference model for depression detection.It encodes the skeleton motion information into images,with features extracted by a convolutional neural network.Based on this,the original model is further extended to probabilistic models following the Bayesian modeling framework.Between-class learning was also used to constrain the distribution of features.Second,considering the possible complementarity of different forms of features,this paper proposes a deep time-frequency classification model for depression detection.It uses a Long Short-Term Memory with a temporal attention mechanism to obtain depth features by modeling the long-term dependence of the 3D human posture in the walking sequence,and then integrates and classifies decisions with the time domain and frequency domain features of joint motion.Experimental results show that the proposed two models have high accuracy and reliability to effectively identify depressed patients.The research results of this paper provide new ideas and methods for the early diagnosis and treatment of depression,and have certain clinical application value.