Cucumber Downy Mildew Disease Prediction Using a CNN-LSTM Approach

Abstract

It is of great significance to develop early prediction technology for controlling downy mildew and promoting cucumber production. In this study, a cucumber downy mildew prediction method was proposed by fusing quantitative disease information and environmental data. Firstly, the number of cucumber downy mildew spores during the experiment was collected by a portable spore catcher, and the proportion of cucumber downy mildew leaf area to all cucumber leaf area was recorded, which was used as the incidence degree of cucumber plants. The environmental data in the greenhouse were monitored and recorded by the weather station in the greenhouse. Environmental data outside the greenhouse were monitored and recorded by a weather station in front of the greenhouse. Then, the influencing factors of cucumber downy mildew were analyzed based on the Pearson correlation coefficient method. The influencing factors of the cucumber downy mildew early warning model in greenhouse were identified. Finally, the CNN-LSTM (Convolutional Neural Network-Long Short-Term Memory) algorithm was used to establish the cucumber downy mildew incidence prediction model. The results showed that the Mean Absolute Error (MAE), Mean Square Error (MSE), Root Mean Square Error (RMSE), and determination coefficient (R²) of the CNN-LSTM network model were 0.069, 0.0098, 0.0991, and 0.9127, respectively. The maximum error between the predicted value and the true value for all test sets was 16.9398%. The minimum error between the predicted value and the true value for all test sets was 0.3413%. The average error between the predicted and true values for all test sets was 6.6478%. The Bland–Altman method was used to analyze the predicted and true values of the test set, and 95.65% of the test set data numbers were within the 95% consistency interval. This work can serve as a foundation for the creation of early prediction models of greenhouse crop airborne diseases.

1. Introduction

In recent years, with the increasing demand for a better life and “vegetable basket” project, China’s vegetable industry has developed rapidly [1,2]. China’s facility cultivation area has reached more than 4.2 million hectares, and the facility cultivation is widely distributed in China, with the largest total area in the world. It is the main support of China’s vegetable basket project and one of the important ways to increase farmers’ income [3,4]. Cucumber is highly favored by people due to its flavorful taste and high nutritious content [5,6]. The cucumber cultivation area accounts for the main part of the facility cultivation area in China [7,8]. Cucumber is inevitably subjected to biological stresses (such as bacterial and fungal diseases) in the growth process, and the temperature and humidity conditions in the greenhouse environment are more conducive to the prevalence and spread of airborne fungal diseases [9,10]. Cucumber downy mildew is caused by Pseudoperonospora [11]. After the disease occurs, it mainly harms the leaves of cucumber plants, as well as the stems and inflorescences of cucumber plants. The disease can occur from the seedling stage to the growth stage of cucumber, especially in the harvest stage, and most leaves of cucumber plants can die within a week or two. The cucumber field is yellow [12,13]. The spread of cucumber downy mildew will worsen year after year as the cultivation area and number of planting cycles grow. In extreme circumstances, it can cause harvest loss, decreased productivity, and yield losses of 20–50% [11,14]. Therefore, the development and application of early diagnosis and prediction technology of cucumber downy mildew are particularly important to guide the prevention and treatment of cucumber downy mildew.

The primary cause for the occurrence and prevalence of cucumber downy mildew is the dispersion of pathogen spores through airflow. The abundance of pathogen spores in the air is directly linked to the presence and prevalence of the illness [15]. Granke et al. [16] conducted a study on the impact of disease spore count on the occurrence and intensity of cucumber downy mildew. They discovered that the number of sporangia in the air, planting time, planting density, temperature, and leaf humidity were positively associated with the severity of cucumber downy mildew. Conversely, solar radiation showed a negative correlation with the severity of cucumber downy mildew. The density of sporangia in the atmosphere is a crucial determinant of the occurrence and intensity of cucumber downy mildew. Xu et al. [17] showed that the number of spores was highly significantly positively correlated with temperature and light intensity (p ≤ 0.01), and negatively correlated with relative humidity (p ≤ 0.01), and the number of newly infected strawberry fruit with gray mold was highly significantly positively correlated with the number of conidia captured on the day before 7 days. In recent years, with the development of deep learning and machine learning technology, data-based disease early warning models have been widely developed and applied. Neufeld et al. [18] used night temperature, daytime temperature, cumulative hours of daily relative humidity >80%, and daily leaf wetting time as input factors to construct a model to predict the risk of downy mildew infection. Zhang et al. [19] introduced a method for predicting wheat stripe rust using a knowledge graph and a bidirectional long short-term memory network. The results show that a knowledge graph enriches the semantic information described in disease prediction, improves the ability of a bidirectional long short-term memory network to extract high-level disease prediction features, and thus improves the accuracy of disease prediction. El Jarroudi et al. [20] v. Wang et al. [21] established a ricE bLast simuLation model based on the ELLE process to reproduce the dynamic changes in the severity and incidence of rice blast in leaves and ears, which provided a tool for seasonal monitoring and scenario assessment. Chelal et al. [22] established a model of interaction between tomato early blighting epidemic and growth dynamics. The model was fitted to data from greenhouse experiments in which tomato plants were inoculated with lycopene three times, and the model parameters were estimated. The existing early diagnosis methods only detect the disease after the plant is infected with the disease, and the detection after the onset of the disease has actually delayed the best period of prevention and control, which often makes the prevention and control of the disease in the passive. On the other hand, the establishment of a disease prediction model mainly relies on the relationship between environmental factors and disease results. In fact, the occurrence of crop diseases is not only related to environmental conditions such as temperature and humidity, but also related to the growth and transmission of the fungal spores of aerial diseases of crops. Therefore, only the spore information of airborne diseases and environmental information such as temperature and humidity can be fused to establish a more accurate disease prediction model.

Therefore, in order to realize the prediction of cucumber downy mildew in greenhouse, this paper proposed a cucumber downy mildew prediction method based on the CNN-LSTM algorithm. The CNN was used to mine the potential information relationship between the input factors of the model. The LSTM was used to solve the problem of predicting time series. Firstly, the incidence data of cucumber downy mildew and greenhouse environment data were collected, and the influencing factors of cucumber downy mildew were analyzed. Then, the influencing factors of the early warning model of cucumber downy mildew in greenhouse are determined. Finally, this study used the greenhouse environmental data during the greenhouse cucumber planting period, the number of cucumber downy mildew spores in the air, and the incidence data of cucumber downy mildew, and used the CNN-LSTM model to find the internal rules between greenhouse environmental factors and the incidence of cucumber downy mildew, and the early warning model of cucumber downy mildew in greenhouse was established. This study can provide a basis for the development of early prediction models of greenhouse crop airborne diseases, and is of great significance for the timely guidance of disease prevention and avoiding the occurrence of catastrophic damage in greenhouse production.

2. Materials and Methods

2.1. Greenhouse Facilities and Experiments

The experiments were conducted from 6 July to 5 August 2020, from 31 August to 30 September 2020, from 1 April to 1 May 2021, and from 15 October to 14 November 2021 in the Venlo greenhouse of Jiangsu University. The greenhouse is oriented east-west, with a top height of 4.73 m, a shoulder height of 4.0 m, a span of 3.2 m per span, and a length of 45 m. The greenhouse has a thermal insulation screen, using pipe heating, a sunshade net, and a wet curtain fan and other cooling equipment, ensuring environmental regulation using a computer automatic control system control.

The cucumber cultivar tested was “Jinyou No.1” (developed by the Institute of Tianjin Academy of Agricultural Sciences). The nutrient soil is put in a hole tray, watered and wetted, and cucumber seeds are put in a hole tray filled with the nutrient soil for cultivation. When the cucumber seedlings grow to “two leaves and one heart”, the cucumber seedlings with good growth and a similar state are removed from the cavity plate and transplanted into the flowerpot with a volume of 10 L. Perlite was used to cultivate cucumber plants. The day before transplanting the cucumber seedlings, the perlite was poured into the pot, and the perlite was rinsed with tap water to reduce the influence of the dust contained in perlite on the growth of cucumber seedlings. The cultivation of cucumber plants is shown in Figure 1. Two rows of cucumber plants were cultivated in each experiment, with 30 plant samples per row. Conventional management is adopted, and nutrient solution irrigation is adopted after transplanting, and the nutrient solution formula is Yazaki nutrient solution [23]. In order to ensure the continuous supply of nutrient solution during the cultivation of cucumber plants, the cucumber plants are watered with nutrient solution from 8:00 to 9:00 every morning, 400 mL each time. Conventional management is adopted, and nutrient solution is used for irrigation after transplanting. In order to ensure the continuous supply of nutrient solution during the cultivation of cucumber plants, the cucumber plants are watered with nutrient solution from 8:00 to 9:00 every morning, 400 mL each time. In Figure 2, cucumber downy mildew spores in the air were collected during the experiment using a portable spore capture instrument (OK-BZ1, Zhengzhou Oukeqi Instrument Manufacturing Co., Ltd., Zhengzhou, China). The disease degree of downy mildew in cucumber was recorded.

The weather station in the greenhouse (Figure 3a) monitored and recorded the environmental data within the greenhouse. The weather station located in front of the greenhouse (Figure 3b) was used to monitor and collect environmental data outside the greenhouse. Environmental information data were recorded every hour. The collected environmental information mainly included the average temperature and average humidity inside the greenhouse, the average temperature and average humidity outside the greenhouse, and the rainfall outside the greenhouse.

2.2. Model Factors and Data Acquisition

2.2.1. Model Input Factor

The incidence and severity of cucumber downy mildew in greenhouses are not only closely related to the temperature and humidity of the growing environment, but also related to the pathogenic factor downy mildew of cucumber. Although the cucumbers planted in the greenhouse are relatively isolated from the environment outside the greenhouse, the environment outside the greenhouse (such as temperature, humidity, and rainfall) will also affect the environment inside the greenhouse. Therefore, considering the actual situation of the experiment, this study determined the main factors affecting the occurrence of cucumber downy mildew [24,25]. The input factors of the prediction model included the quantity of cucumber downy mildew spores, the severity of cucumber downy mildew, the average temperature and humidity inside the greenhouse, the average temperature and humidity outside the greenhouse, and the amount of rainfall outside the greenhouse.

2.2.2. Data Acquisition

(1)

Disease data

In this work, the number of pathogenic spores in the air and the disease degree of cucumber plants during cucumber planting were collected, and the data were gathered in the Venlo greenhouse of Jiangsu University. In order to ensure the robustness of findings across different time frames, the collection of experimental data was divided into four batches, from 6 July to 5 August 2020, from 31 August to 30 September 2020, from 1 April to 1 May 2021, and from 15 October to 14 November 2021. The data collected on the number of pathogen spores in the air during cucumber planting and the incidence degree of cucumber plants are shown in Figure 4.

(2)

Environmental data

A little weather station, both inside and outside the greenhouse, is set up to gather environmental information data hourly. The most common pieces of environmental data gathered include typical indoor and outdoor temperatures and humidity levels as well as precipitation data from the surrounding area. Figure 5 shows the acquired environmental data, both inside and outside the greenhouse, for the experiment.

2.2.3. Data Processing

This study included seven historical days’ worth of data, including greenhouse temperature, relative humidity, outdoor temperature, relative humidity, and rainfall outside the greenhouse [17,18,20]. Measurements are taken both inside and outside the greenhouse for the following seven days (from day 8 to day 14). This includes the following: temperature inside the greenhouse, relative humidity inside, temperature outside the greenhouse, relative humidity outside, and rainfall outside the greenhouse. The number of spores collected in the greenhouse and the disease degree of cucumber plants were used to predict the cucumber downy mildew disease degree on the 8th day in the greenhouse. The environmental data collected inside and outside the greenhouse are hourly data, and the data need to be averaged. The environmental data collected inside and outside the greenhouse are hourly data, so it is necessary to average the data. Therefore, the data collected 24 times a day are averaged as the environmental data of the day. In addition, the collected data contain date and time information, which needs to be consolidated and renumbered. The processed data are shown in

Table 1. Average value.

Number	Model Factor
	TempR (°C)	RHR (RH)	TempT (°C)	RHT (RH)	Rain (mm)	TempTF (°C)	RHTF (RH)	RainF (mm)	Spore (units)	Disease (%)
1	24.95	91.1	23.78	96.5	0	23.51	87.3	0	15	0
2	27.23	88.5	25.79	93.3	0	23.06	95.9	1.5	8	0
3	27.63	82.7	26.2	88.4	0	21.71	90.2	0	4	0
4	25.87	82.6	24.33	90	0	22.9	86.2	0	2	0
5	25.45	84.5	24.25	88.1	0	23.61	96.7	5.3	1	0
…	…	…	…	…	…	…	…	…	…	…
119	18.56	74.92	17.62	49.96	0	19.64	46.64	0	23	85.5
120	21.83	71.56	21.69	54.48	0	68.79	68.79	0	13	88.7

Note: TempR is the temperature inside the greenhouse, RHR is the relative humidity inside the greenhouse, TempT is the temperature outside the greenhouse, RHT is the relative humidity outside the greenhouse, Rain is the rainfall outside the greenhouse, TempTF is the future temperature outside the greenhouse, RHTF is the future relative humidity outside the greenhouse, and TEMptF is the future relative humidity outside the greenhouse. RainF is the future rainfall outside the greenhouse, Spore is the number of spores collected inside the greenhouse, and Disease is the disease degree of the cucumber plant.

.

The input variables of the model were derived from the 10 components listed in Table 1. To mitigate the mistakes arising from variations in orders of magnitude and units of the model input factors, it is imperative to normalize these parameters before inputting them into the model. The calculation method is presented in Formula (1) after normalization.

$$x_i^{*}=\frac{x_i-x_{\min }}{x_{\max }-x_{\min }}$$

(1)

where $x_{\max }$ represents the highest value of the associated input factor in the sample data. $x_{\min }$ represents the lowest value of the associated input factor in the sample data. $x_i$ represents the input factor of the model in the sample data. $x_i^{*}$ is expressed as the normalized value of the corresponding model input factor in the sample data.

2.2.4. Data Analysis

In order to establish a better early warning model of greenhouse crop airborne diseases, it is important to analyze the input factors of the model. In this work, the Pearson correlation coefficient was used to measure the relationship between the input factors of the model. Figure 6 illustrates the correlation between the input variables of the model.

It can be seen from Figure 6 that although there is a certain relationship between the disease degree of greenhouse crop airborne diseases and various input factor of the model, the linear relationship is weak or non-existent. Therefore, the traditional mathematical modeling method may not be suitable for establishing the early warning model of greenhouse crop airborne diseases.

2.3. Prediction Model

The correlation between cucumber downy mildew incidence and greenhouse environmental factors is intricate and non-linear. Existing mathematical models struggle to establish a precise prediction model for cucumber downy mildew incidence. However, machine learning or deep learning-based models can effectively extract the underlying patterns between greenhouse environmental factors and cucumber downy mildew incidence, surpassing other mathematical models. Therefore, this study used the greenhouse environmental data during the greenhouse cucumber planting period, the number of cucumber downy mildew spores in the air, and the incidence data of cucumber downy mildew, and used the CNN-LSTM model to find the internal rules between greenhouse environmental factors and the incidence of cucumber downy mildew, and the early warning model of cucumber downy mildew in greenhouse was established.

2.3.1. CNN

According to the analysis of Figure 6, although there is a certain relationship between the input factors of the model, the linear relationship is weak, and the relationship between the input factors of the model is also complicated. Therefore, this study uses the Convolutional Neural Network (CNN) to mine the potential information relationship between the input factors of the model. The CNN can use local connection and weight sharing to obtain the effective feature relationship between each input factor of the model through the convolutional layer and pooling layer, and then automatically extract the effective features of each input factor of the model and form a feature vector [26,27].

2.3.2. LSTM

The Long Short-Term Memory network (LSTM) is a sort of time recurrent neural network improved from a recurrent neural network, which can be used to predict time series problems [28]. There are three gates in a cell of the LSTM, which are called the input gate, forget gate, and output gate [29]. The input gate is used to determine the value that will be updated. When a piece of information enters the LSTM network, it can be judged according to the rules. Only information that conforms to the authentication of the algorithm is left, and information that does not conform is forgotten through the forget gate [30].

2.3.3. CNN-LSTM

In the establishment of the greenhouse crop airborne disease early warning model, because the input factors of the model are time series, and the relationship between the input factors is complex, there may be a certain coupling relationship. Therefore, this study combines the characteristics of the CNN and LSTM. The CNN is used to obtain the effective feature relationship between the input factor of the model and forms the feature vector, and the LSTM is used to process the time series of input factors [31,32]. The structure of the CNN-LSTM network is shown in Figure 7. The incidence of cucumber downy mildew is related to the average temperature, the average humidity, the average temperature outside the greenhouse, the average humidity outside the greenhouse, the rainfall outside the greenhouse, the average temperature outside the greenhouse, the average humidity outside the greenhouse, the rainfall outside the greenhouse, and the number of cucumber downy mildew spores in the greenhouse. A total of 10 model input factors are connected into a vector to form a new input factor with a time series. The input factors are input into the CNN network model and processed by the convolution layer and aggregation layer to extract features, and then the extracted feature vectors are input into the LSTM network model.

3. Results and Analysis

3.1. Evaluation Index

In this study, the Root Mean Square Error (RMSE), Mean Absolute Error (MAE), Mean Square Error (MSE), and determination coefficient (R²) were used as the performance evaluation indicators of the prediction model. The RMSE, MAE, MSE, and R² are calculated as follows.

$$RMSE=\sqrt{\frac{1}{\mathrm{n}}{\displaystyle \sum _{i=1}^n{\left(y_i-\stackrel{-}{y_i}\right)}^2}}$$

(2)

$$MAE=\frac{1}{\mathrm{n}}{\displaystyle \sum _{i=1}^n\left|y_i-\stackrel{-}{y_i}\right|}$$

(3)

$$MSE=\frac{1}{\mathrm{n}}{{\displaystyle \sum _{i=1}^n\left(y_i-\stackrel{-}{y_i}\right)}}^2$$

(4)

$$R^2=\frac{{\displaystyle {\sum }_{i=1}^n{\left(\stackrel{-}{y_i}-\stackrel{-}{y}\right)}^2}}{{\displaystyle {\sum }_{i=1}^n{\left(y_i-\stackrel{-}{y}\right)}^2}}$$

(5)

where $y_i$ is used to represent the true value of the degree of disease in cucumber plants. $\stackrel{-}{y_i}$ is used to indicate the predicted value of the incidence degree of cucumber plants. $\stackrel{-}{y}$ is used to represent the central value of the degree of disease in cucumber plants. $n$ is used to denote the number of sample test sets.

3.2. Prediction Results

In this study, when the CNN, LSTM, and CNN-LSTM network models were used to predict the disease degree of cucumber downy mildew in greenhouse, the inputs of the network included 10 variables: cucumber downy mildew disease degree, average temperature inside the greenhouse, average humidity inside the greenhouse, average temperature outside the greenhouse, average humidity outside the greenhouse, rainfall outside the greenhouse, average temperature outside the greenhouse, average humidity outside the greenhouse, rainfall outside the greenhouse, and the number of cucumber downy mildew spores in the greenhouse, and the parameters of the first 7 days were used to predict the cucumber downy mildew disease degree on the 8th day. Therefore, the input sample size was 8 × 10, the time_step was set to 8, the number of iterations was set to 300, and the validation_split was set to 0.2 during network training. The running results of the CNN, LSTM, and CNN-LSTM network models are shown in Figure 8, Figure 9, and Figure 10, respectively.

3.3. Results Statistics and Analysis

In this study, three prediction models, the CNN, LSTM, and CNN-LSTM, were established to predict the disease degree of cucumber downy mildew in greenhouse, and the statistical results of the three network models are shown in

Table 2. Comparison of three network models’ performances.

Model	Evaluation Index
	MAE	MSE	RMSE	R2
CNN	0.0992	0.0190	0.1378	0.8312
LSTM	0.0961	0.0200	0.1416	0.8217
CNN-LSTM	0.0690	0.0098	0.0991	0.9127

. It can be seen from Table 2 that the minimum MAE of the CNN-LSTM network model was 0.069, which was 30.44% and 28.2% lower than that of the CNN and LSTM network models, respectively. The minimum MSE of the CNN-LSTM network model was 0.0098, which was 48.42% and 51% lower than that of the CNN and LSTM network model, respectively. The minimum RMSE of the CNN-LSTM network model was 0.0991, which was 28.08% and 30.01% lower than that of the CNN and LSTM network model, respectively. The maximum R² of the CNN-LSTM network model was 0.9127, which was 9.8% and 11.07% higher than that of the CNN and LSTM network model, respectively. Therefore, the prediction effect of cucumber downy mildew based on the CNN-LSTM model was better than that of the CNN and LSTM model, respectively. This is because when the CNN-LSTM model was used to predict cucumber downy mildew, the CNN-LSTM model was used to not only find the internal law between greenhouse environmental factors and cucumber downy mildew incidence, but also solve the time series prediction problem between greenhouse environmental factors and cucumber downy mildew incidence.

When the CNN-LSTM network model was used to predict the extent of cucumber downy mildew disease in greenhouse, the error results between the predicted value and the true value of the test set are shown in Figure 11. It can be seen from Figure 11 that the maximum error between the predicted value and the true value for all test sets was 16.9398%. The minimum error between the predicted value and the true value for all test sets was 0.3413%. The average error between the predicted and true values for all test sets was 6.6478%. Compared with the prediction method combining the knowledge graph and bidirectional long short-term memory network, it is 0.1422 percentage points higher [19]. In addition, with the increase in the number of the test days, there is an increase in the cucumber downy mildew disease degree in the future days in the actual situation. The error between the predicted value and the true value of the test set fluctuates to some extent, but shows an increasing trend. This is because when predicting the extent of cucumber downy mildew disease in the greenhouse one day in the future, the forecast value of the previous day is used (for example, in predicting the extent of the next day, the forecast value of the first day is used as the model input, and in predicting the extent of the fifth day, the forecast value of the first four days is used as the model input, and so on). As a result, the error between the predicted value of the test set and the real value of the cucumber downy mildew disease degree in the greenhouse will increase with the increase in the future date. Therefore, in this practical application, it is suggested that when using historical experimental data to predict the future degree of greenhouse cucumber downy mildew disease, in order to improve the prediction accuracy, the real value of the model input should be updated in time, so as to avoid the excessive use of the predicted value to re-input the model to predict the future degree of greenhouse cucumber downy mildew disease.

In this study, in order to further illustrate the feasibility of the CNN-LSTM network model proposed, the temperature inside the greenhouse, the relative humidity inside the greenhouse, the temperature outside the greenhouse, the relative humidity outside the greenhouse, and the rainfall outside the greenhouse were used for seven historical days. For the next 7 days (that is, from day 8 to day 14, the data inside and outside the greenhouse are recorded), we included the temperature inside the greenhouse, the relative humidity inside the greenhouse, the temperature outside the greenhouse, the relative humidity outside the greenhouse, and the rainfall outside the greenhouse. The number of spores collected in the greenhouse and the disease degree of cucumber plants were used to predict the cucumber downy mildew disease degree on the 8th day in the greenhouse. The Bland–Altman method was used to analyze the predicted and true values of the test set, and the results are shown in Figure 12.

It can be seen from Figure 12 that 22 data points in the data number of the test set are within the 95% consistency interval, and only 1 data point is inside or outside the 95% consistency interval, and the data points within the 95% consistency interval account for 95.65% of the total data points. Therefore, it can be shown that the predicted value and the real value have good consistency when the CNN-LSTM network model proposed in this study is used to predict the disease degree of cucumber downy mildew in greenhouse. In conclusion, the CNN-LSTM network model proposed in this study has the best performance, can better mine the hidden information between the variables of the model, and better predict the disease degree of greenhouse cucumber downy mildew, which can provide a reliable decision-making basis for the prediction and early warning of greenhouse cucumber downy mildew.

In this study, cucumber downy mildew was taken as the research object, and the regional weather forecast data for the next 7 days, the number of different types of spores of the disease, and the historical environmental information of the temperature and humidity inside and outside the greenhouse were taken as the input quantities, and the expected incidence degree of crop airborne diseases in the greenhouse was taken as the output quantity. An airborne disease prediction model based on the CNN-LSTM algorithm was established. The early and accurate prediction of the corresponding airborne diseases in greenhouse is realized.

In the future, crop incidence information and atmospheric circulation information in surrounding areas can be considered to optimize and improve the proposed cucumber downy mildew early warning model. In addition, more types of airborne disease spores can be studied, and the parameters of the proposed network model can be optimized, so as to improve the disease early warning effect. In this study, a portable spore capture instrument was used to capture airborne disease spores. However, in reality, the detection of disease spores may still have some limitations for farmers. Therefore, it is suggested to develop a disease spore online detection technology and build a big data platform integrating disease spore information and environmental information for the real-time prediction of crop airborne diseases.

4. Conclusions

In this study, a cucumber downy mildew prediction method was proposed by fusing quantitative disease information and environmental data. The CNN-LSTM algorithm was used to establish the cucumber downy mildew incidence prediction model. The results show that the MAE, MSE, RMSE, and R² of the CNN-LSTM network model were 0.069, 0.0098, 0.0991, and 0.9127, respectively. The maximum error between the predicted value and the true value for all test sets was 16.9398%. The minimum error between the predicted value and the true value for all test sets was 0.3413%. The average error between the predicted and true values for all test sets was 6.6478%. The Bland–Altman method was used to analyze the predicted and true values of the test set, and 95.65% of the test set data numbers were within the 95% consistency interval. Moreover, 22 data points in the data number of the test set are within the 95% consistency interval, and only 1 data point is inside or outside the 95% consistency interval, and the data points within the 95% consistency interval account for 95.65% of the total data points. Therefore, it can be shown that the predicted value and the real value have good consistency when the CNN-LSTM network model proposed in this study is used to predict the disease degree of cucumber downy mildew in greenhouse. In addition, it is suggested that more types of airborne disease spores should be studied in the future and the parameters of the proposed network model should be optimized to improve the universality of early warning.