Study on the Diagnosis of Light Environmental Impact on Cyclocybe chaxingu Using Improved ResNet18 Network

To investigate the effects of light environment on the growth and development of Cyclocybe chaxingu, three different light qualities, green light, red light, and blue light, were set up, and cultivation experiments were conducted under the same light intensity and illumination time for each light quality. Visible light images of Cyclocybe chaxingu were obtained during the growth and maturity phases, and the image data were enhanced by flipping transformation and noise disturbance. The Convolutional Block Attention Module (CBAM) module was added to each residual block of the ResNet18 convolutional neural network (CNN), and an improved ResNet18 network was constructed to diagnose the effects of light environment on Cyclocybe chaxingu image data. The results showed that: 1) in the growth period, the recognition accuracy of the improved ResNet18 model was 99.93%, the model training time was about 3519 s, and the model size was about 1/11 of the VGG11 model. 2) In the maturity period, the accuracy of the improved ResNet18 network was 99.91%, which was more than 8% higher than that of the MobileNet. The above results show that the diagnostic method of the light environment impact on Cyclocybe chaxingu by adding CBAM into ResNet18 residual block is feasible, and can effectively diagnose and identify the light environment of Cyclocybe chaxingu during the growing and maturing periods, and the model has good robustness, which can provide light regulation and optimization strategies for Cyclocybe chaxingu cultivation.