TABLE 1.

Confusion matrix. A mosquito image recognition system was developed using the SqueezeNet neural network, and transfer learning was applied using the mosquito datasets, which were generated by fixing a mosquito specimen on an insect pin and rotating it 360° along the Z-axis, while the camera rotated along the X-axis. LED brightness was adjusted to simulate mosquitoes under different sunlight and other light source conditions. The training database contained images of mosquitoes of different species, sexes, and ages as well as images with different brightness levels and taken at different angles. Through the live mosquito dynamic image capturing device, pictures of mosquitoes in flight and in other positions were taken, increasing the diversity of the training data.

Predicted class
		Ae. aegypti (%)	Cx. quinquefasciatus (%)	Empty (%)
True Class	Ae. aegypti	91.57	1.43	6.99
	Cx. quinquefasciatus	1.25	89.29	9.45
	Empty	4.89	4.56	90.54