Behavior estimation of an insect based on flight muscle electromyograms using the support vector machine

In this study, we propose a behavior estimation method using muscle potential for quantitatively analyzing the behavior pattern of an insect. In previous behavioral analysis studies, it is possible that the analysis result is different from the behavior pattern that insects intended originally because we define behavior patterns statistically using the amount of movement and velocity calculated from trajectory. Therefore, we analyze the behavior state transition at high resolution from the command neuron side signal. In other words, estimating behavior from muscle potential information can analyze behavior more quantitatively than analyzing behavior from trajectory. Thus, it is possible to properly elucidate the behavioral expression mechanism or the behavior selection process. Moreover, it allows us to understand the behavior pattern that insects actually selected. We focus on the odor source searching behavior of silkworm moths, and we estimate three types of behavior: forward moving, left turning, and right turning using a Support Vector Machine (SVM) approach. We employ flight muscle electromyograms (EMGs) whose motor commands were assumed to be generated in the same ganglion as the behavior that was estimated. Flight muscle EMGs of silkworm moths were measured in a free walking state by using a locomotion-measuring instrument named 3 DOF servosphere. We carried out simultaneous measurements on nine silkworm moth individuals, estimated the behavior patterns from the experimental data, and demonstrated the effectiveness of the method.