Project Tag: deep learning

Analysis of the possibility of automatic recognition of the…

This work presents proposals of measures and algorithms helpful in the automatic analysis of complex social behavior of laboratory animals. It focuses on:

  1. accurate segmentation of animals during physical contact, which often involves significant covering of objects,
  2. detection of aggressive behavior on the basis of a thermographic image of a trace of saliva left on the rodent’s fur by another individual.

The possibility of identifying individuals and taking measurements also during the object connection provide data important from the point of view of the analysis of social behavior. For this purpose, the paper proposes a 3-step algorithm for segmentation of connected objects, depending on the degree of animal contact.

 

The trace of saliva left on the fur of a rodent by another individual, in turn, is probably the only discernible evidence of an extremely aggressive interaction between animals (bites or aggressive cleaning). The paper proposes parameters suitable for the description and identification of the analyzed traces, the characteristics of the actual traces were analyzed and the influence of the parameters of detectors and traces on the detection results on simulation images was examined. Based on the observations, a model of temperature changes of the trace was developed, and then the analysis of these traces was performed. It has also been shown that it is possible to predict the temperature of the saliva trace when it is not visible, even for a small number of observations.