Plant Growth Situation Monitoring Based On Channel Information Attention Mechanism

In the process of plant growth,it is very necessary to timely and effectively monitor the growth situation of plants.Therefore,the timely detection and treatment of plant diseases and the effective monitoring of plant moisture are the prerequisites for good plant growth.At present,traditional machine learning and deep learning methods are mainly used to identify plant diseases and moisture.The traditional machine learning method has a complicated data preprocessing process,which is not universal,and requires high data distribution.Deep learning method is now the mainstream method,but the existing deep learning methods ignore the problem of the loss of underlying features and information overload caused by the deepening of network layers.Aiming at the above problems,this paper adopts the improved deep learning method to monitor and identify the plant growth situation from the two aspects of plant diseases and moisture.(1)Aiming at the problem of plant disease recognition,a plant leaf disease recognition model was proposed by fusing the attention network of channel information.Firstly,in order to avoid the loss of underlying features,a basic network based on residual structure is designed to automatically extract the features from the shallow to the deep level.Secondly,in order to give full play to the importance of different features,attention mechanism is introduced,which aims to amplify effective features and suppress useless information at the same time.Finally,the channel information attention network is fused in different layers of the basic network.Experimental results show that the classification accuracy of the fused network is 4.64% higher than that of the basic network,and the accuracy rate of the fused model is 4.23% higher than the network model with the highest recognition rate among the VGG?11,VGG?16,Res Net18 and TCCNN models,and the complexity of the fused network is lower than that of other comparative networks.In general,the fusion channel information attention network can effectively improve the recognition of plant diseases.(2)Aiming at the problem of moisture recognition in plants,the idea of transfer was introduced to realize the recognition of moisture in leaves by the fusion channel information attention network model.Firstly,the experimental paradigm was designed and the data were collected.The data were augmented and equalized by the methods of data enhancement and random erasation.Secondly,the above model was used to retrain the moisture data.Meanwhile,in order to further verify the generalization ability of the model,the pre-training network based on VGG?11 and Res Net18 was used to compare with the above model.Finally,VGG?11 and Res Net18 were used as feature extractors and CNN fine-tuning methods to train and verify the model.Experimental results show that compared with the two methods of the transfer model,the classification accuracy of the fusion channel information attention network model is higher.The fusion channel information attention network model proposed for plant disease recognition can achieve better recognition effect in the shallow layer of the network,and the complexity of the model is low.At the same time,the model was applied to the recognition of plant moisture by using the transfer idea,and the final recognition accuracy was better than that of VGG and Res Net18 networks,which verified the feasibility and effectiveness of the fusion channel information attention network.