Plant growth prediction for lettuces using photosynthetic information

[OS-8-04] The XX CIGR World Congress 2022 Plant growth prediction for lettuces using photosynthetic information Yuki Ito 1, Yuya Hosoda 2, Hitoshi Goto 1, 2, 3, Seitaro Toda 4, Kotaro Takayama 4 1 Department of Computer Science and Engineering, Toyohashi University 2 Center for IT-based Education, Toyohashi University 3 Information and Media Center, Toyohashi University 4 Department of Mechanical Engineering, Toyohashi University of Technology, of Technology, of Technology, of Technology, Japan Japan Japan Japan

Outline Goal ✓ To reduce running costs for the plant factory by optimizing running plans. ✓ To predict the lettuce weight at shipping stage based on growth data at the seedling stage. Method ✓ Obtains the projection area and edge length from chlorophyll fluorescence images as growth data. ✓ Predicts the lettuce weight using curve fitting and multiple linear regression analysis. Results ✓ The proposed method predicts the lettuce weight on the 20th day after planting in the growing rack using growth data up to the 10th day. 2

3 Background Severe weather Chemical fertilizer Cycle Pump Container Tank Water pollution Open-field culture Filter Liquid fertilizer Plant factory ✓ Has low running costs using nature. ✓ Has high running costs by controlling systems. ✓ Determines the yield due to weather conditions. ✓ Provides stable yields year-round in the container. ✓ Faces challenges to environmental protection. ✓ Reduces water pollution through hydroponics.

Goal | Optimization for Running Plans Growth Prediction Predicts of shipping stage conditions based on growth data at the seedling stage. Seedling stage Shipping Stage ✓ Shipping plan Growth Prediction Seedling stage Shipping Stage Growth Prediction Optimization ✓ growing environment 4

Related works | Plant Growth Prediction Weight Prediction Shape Prediction Predicts the shape of leaves in the future using an encoder-decoder architecture. Predicts the lettuce weight in the future based on the leaf movement. Leaf movement analysis using optical flow Encoder Decoder Previous data Sakurai, S., Uchiyama, H., Shimada, A., Taniguchi, R., “Plant Growth Prediction using Convolutional LSTM”, In Proc. VISIGRAPP, 2019. ✓ Growth & Stagnation Future data ✓ Direction of extension Nagano, S., Moriyuki, S., Wakamori, K., Mineno, H., Fukuda, H., “Leaf-movement -based growth prediction model using optical flow analysis and machine learning in plant factory”, Front. Plant Sci., 10, 2019. 5

6 Proposed method Chlorophyll fluorescence image 1st-N th Days Projection area Projection Area Observed 1 Curve Fitting Future 10 Multiple Linear Regression Analysis 20 Date Edge Length Edge Length Future Observed 1 Curve Fitting 10 Date 20 20th day

7 Method | Image Data Obstacle Brightness Difference RGB Color image ✓ Observes the light that was reflected from objects. ✓ Requires pre-processing to detect plants. Without Obstacle Less Brightness Difference Chlorophyll fluorescence image ✓ Observes the absorbed light energy that was not used in photosynthesis. ✓ Detects plants without pre-processing.

Method | Feature for Projection Area Projection Area ✓ Represents the size of the plant as seen from above. ✓ Is calculated using Otsu’s method that automatically performs thresholding. ✓ Takes the sum with the previous day’s one because of including the shadowed area of the leaves. Shadowed area Projection area Observed value Chlorophyll fluorescence image Binary image 8

9 Method | Feature for Edge Length Edge Length ✓ Represents the overlapping of leaves as seen from above. ✓ Is calculated using the Canny edge detector that detects the edge using thinning process. Edge length Observed value Chlorophyll fluorescence image Thinning image

10 Method | Curve Fitting 𝑦𝑃 𝑥 = 1+ 𝐾 𝐾 −1 𝑦𝑃 (1) 𝑒 −𝑎(𝑥−1) 𝑦P ｜Projection area 𝑥 ｜Date Observed value 𝑦E ｜Edge Length Approximate 𝑥 ｜Date function Observed value Approximate function

11 Method | Weight Prediction Multiple Linear Regression Analysis Weight Predicted Projection Area Predicted Edge Length

12 Experiment Environment Material Temperature 24-28 degrees ✓ 45 Lettuce Plants (were planted in the growing rack 14 days after sowing) Height 30 cm ✓ Cultivation for 20 days (considered the seedling stage up to 𝑁-th day after planting and the shipping stage on the 20th day) ✓ Closs Validation (k=5) (Train | 36 plants / Test | 9 Plants) Frillice lettuce (Chile)

Results 1/2 (Observed Date Train ) Test 13

14 Results 2/2 Bad More observed values make stable curve fitting to capture future values. The longer days until the shipping stage can optimize the running plan. 22.7 Good

Conclusion & Future Works Method ✓ Obtains projected area and edge length from chlorophyll fluorescence images as growth data. ✓ Predicts the weight using curve fitting and multiple regression Linear analysis. Results ✓ The proposed method predicts the lettuce weight on the 20th day after planting using growth data up to the 10th day after planting. Future ✓ Improves the feature extraction using curve fitting with multiple functions. ✓ Increases plant samples to test the effectiveness of the proposed method. 15