Irrigation Strategy Of Greenhouse Tomato In Vegetative Growth Stage Based On Artificial Intelligence

Irrigation is the only source of water for crops in greenhouse.The time and volume of irrigation are the key to achieve accurate irrigation.In order to overcome the deficiency that existing irrigation strategies can not dynamically adjust irrigation timing and amount according to crop water demand in a day,an irrigation strategy of greenhouse tomato in vegetative growth stage based artificial intelligence was proposed.In this method,artificial intelligence technology was used to detect the water status of tomatoes,predict the plant transpiration.This method dynamically adjust the time and volume of irrigation to achieve accurate irrigation of crops.According to the proposed irrigation strategy for greenhouse,an intelligent irrigation system was developed.The specific research contents are as follows:(1)A detection model of water status of greenhouse tomato based on multimodal deep learning was proposed.Through the experiment,it was found that the upper leaves of tomato plants are first part to show the symptoms of water deficiencyand the top leaf of tomato was determined as the detection site of plant water status.Through irrigation with different proportions of water,tomato samples with five different water states,namely,severe irrigation deficit,slight irrigation deficit,suitable irrigation,slight irrigation excess and severe irrigation excess,were cultivated.A total of 26 detection models of tomato water status based on single mode deep learning and multimodal deep learning were constructed.RGB images,depth images and near infrared images of the upper leaves of tomato samples with different moisture states were used for training.The features of the images were extracted by VGG-16 and Res Net-50.The water status detection model of tomato was constructed by using one,two and three of RGB images,depth images and near-infrared images respectively.The results showed that the minimum accuracy of tomato water state detection networks using one image,two images and three images were respectively 88.97%,93.09%and 98.89%,and the maximum accuracy were respectively 93.09%,96.90%and 99.19%.The multimodal deep learning network was superior to the single-modal deep learning network in the detection of tomato water status.There was a similar accuracy between VGG-16 and Resnet-50 in tomato water status detection,but the size of of weight parameter file of VGG-16 is about 5 times that of Res Net-5,so Res Net-50 was selected.The accuracy of detection model of tomato water state using three images reached 99.19%.(2)The prediction model of greenhouse environmental parameters and the calculation model of tomato transpiration based on long short-term memory(LSTM)were constructed.Taking the greenhouse environment of the past 24 hours as the input and the greenhouse environment of the next hour as output,the prediction model of greenhouse environment parameters based on LSTM was established.The determination coefficients of temperature,relative humidity and photosynthetically active radiation were 0.9992,0.9947 and 0.9964,respectively.The root mean square errors were respectively 0.3°C,1.50%and 9.68μmol/(m~2·s).The calculation model of tomato evapotranspiration takes indoor air temperature,relative humidity,photosynthetically active radiation,planting date,current date and current time as the input of the model,and evapotranspiration as the output.The root mean square error of the calculated values of evapotranspiration was 0.4 g,and the determination coefficient was 0.9812.The predicted values of greenhouse environmental parameters were input into the tomato transpiration calculation model,and the predicted values of tomato transpiration in the next period could be obtained.The results showed that the root mean square error of the predicted value of tomato evapotranspiration obtained by this method was 0.5 g,and the determination coefficient reached 0.9396.(3)The irrigation strategy of greenhouse tomato in the vegetative growth stage based on artificial intelligence was formulated,and the corresponding irrigation system was developed.The irrigation strategy is as follows:The time of irrigation is adjusted according to the water status of tomato detected by the detection model of water status of greenhouse tomato based on multimodal deep learning.If the tomato is found to be in a state of water shortage,the irrigation shall be carried out immediately.The irrigation volume is determined by the greenhouse environment prediction model based on LSTM and the tomato transpiration calculation model to realize the dynamic adjustment of irrigation amount in a day.The irrigation strategy was input into the Aquacrop tomato growth model for simulation.Compared with the irrigation method based on matrix moisture content and the irrigation method based on water consumption in the previous period,the dry matter water use efficiency increased by6.5%and 17.3%,respectively.The control system of automatic irrigation machine was developed,which included tomato image acquisition module,greenhouse environmental parameter acquisition module and irrigation decision module,and the irrigation effect was tested and verified.The results showed that the average daily increase of plant height was 1.01 cm and 0.98 cm,respectively,and the average daily increase of stem diameter was 0.041 mm and 0.042 mm,respectively,during the vegetative growth period of tomato under irrigation strategy and timing and quantitative irrigation strategy.The plants growth was similar between the two strategies and the irrigation strategy saved 17.82%water compared with timing and quantitative irrigation strategy.In conclusion,the detection model of tomato water status,the prediction model of greenhouse environmental parameters and the prediction model of tomato transpiration were constructed,which can detect water status of tomato in time and predict the transpiration of tomato.The time and volume of irrigation can be dynamically adjusted.This study realizes the precise irrigation of crops,saves irrigation water and reduces the waste of water resources.