In this study, a minimum of 500 thermal images were taken of the queen bee, worker bees, and drones using a thermal camera. For this purpose, a box with dimensions of 20×20 cm 2 and a variable height of 15 to 50 cm, along with the XY-TR01 temperature and humidity control module, was used. The data were processed in the InfRec Analyzer NS9500 Standard thermal analysis software, and then the deep learning model was developed for bee type classification. Examination of thermal images showed that the average body surface temperature of the queen is approximately 35.53°C, the worker is around 40°C, and the drone is about 39.5°C. Two tools, a histogram, and a co-occurrence matrix, were used to extract features from the thermal images and process them. The extracted features were used to train the artificial neural network with different activation functions. Also, thermographic images acquired from bees were used to develop a convolutional neural network. The results of identifying the type of honeybee based on thermal images using an artificial neural network showed that the best model for achieving the highest accuracy in determining the type of bee is the neural network trained with the Trainscg function. The final accuracy of training and evaluation of this network was observed to be 98.6% and 87.3%, respectively. Additionally, a Perceptron neural network (PNN) and a Convolutional Neural Network (CNN) were used, and the results showed that this model could classify all honeybees' thermal images into the correct classes with 100% accuracy. • Determining the distribution of body surface temperature of honey bees using thermal imaging. • Minimum and maximum body surface temperatures were for queen and male bees. • Successful classification of honey bee type using machine learning models. • Improving accuracy in honey bee type classification using CNN • Achieving 100% accuracy in honey bee classification using CNN model
Derakhshi et al. (Sun,) studied this question.