LED Plant Intelligent Lighting System Based On Deep Learning

The industrialization of plant planting is the mainstream direction of modern agricultural development.For plant factories,artificial lighting is a key link in plant planting and the main means of affecting plant growth.Accurate lighting can enable plants to achieve optimal growth effects in long environments.Being able to accurately identify the growth status of plants and match the corresponding lighting conditions based on the growth status is a prerequisite for achieving precise lighting in plant factories.Applying artificial intelligence technology to achieve plant recognition and classification can provide a regulatory basis for intelligent lighting.Intelligent lighting is the development direction of modern agricultural plant planting technology.Applying the classification recognition neural network in deep learning,in response to the problem of the lack of continuous growth image data within a cycle in the publicly available image dataset of plants,the growth images of multiple sets of tomatoes under different conditions within a cycle were collected.Finally,the collected tomato plant image data was processed and enhanced,and a convolutional neural network with outstanding performance was selected in a series of network models based on plant characteristics,Including Efficient Net,Reg Net,Shuffle Net,and Dense Net,four neural networks were trained and tested based on the same training and validation sets.In the end,the recognition accuracy of each neural network on the validation set reached 96.6%,95.6%,95.5%,and 96.7%,respectively.Based on the Efficient Net model,which performs well in terms of stability and accuracy,algorithm optimization was carried out to improve the accuracy of the neural network model from 92.4% to97.5%,resulting in a high accuracy plant recognition neural network model.Improving the interpretability of neural networks can enhance the mastery and understanding of their internal working mechanisms.Visualize the convolutional kernels and feature maps in the neural network,and display the training process of the neural network in the form of images.Apply the Grad-CAM algorithm to the output results of the neural network to obtain a thermal map of the input image.When the neural network generates different predictive values for the input image,the weight parameter matrix of the full connection layer is derived,and the difference between the two groups of weight parameters is corrected.The new gradient information of the feature layer is calculated through the updated weight parameters,and the counter fact conceptual interpretation algorithm is implemented to screen out the spurious relationship factors with the target task,thus improving the accuracy of the neural network.Finally,a plant intelligent lighting system was designed and developed based on microcontroller controlled PWM wave to achieve linear control of LED lights.Use Raspberry Pi to collect plant growth status images,analyze and match the corresponding lighting requirements based on the information identified from the images,remotely control the LED plant intelligent lighting system,so that the lighting source can be adjusted according to the different growth stages of the plant and the parameters of the light source,thereby achieving precise illumination of the plant growth process.