Ensemble of Machine Learning Algorithms for Rice Grain Yield Prediction Using UAV-Based Remote Sensing

Sarkar Tapash Kumar,Roy Dilip Kumar,강예성,전새롬,박준우,류찬석 한국농업기계학회 2024 바이오시스템공학 Vol.49 No.1

Purpose Accurately estimating rice yield before harvesting is crucial for eff ective crop management, food trade assessment, and national food policy planning to ensure food security. Remotely sensed spectral information such as vegetation index (VI)-based approaches for yield prediction are adequate during mid-stage growth but not during ripening due to leaf senescence, canopy coverage, panicle abundance, and other factors. To fi ll this research gap, this study aims to predict rice yield during ripening stage using an ensemble of machine learning (ML) algorithms. Methods A fi xed-wing unmanned aerial vehicle (UAV) was employed to acquire spectral features from red-green-blue, nearinfrared, and red-edge images. In this study, we utilized state-of-the-art ML-based algorithms, such long short term memory (LSTM), bi-directional LSTM (Bi-LSTM), Gaussian process regression (GPR), fuzzy inference system (FIS), adaptive neuro FIS (ANFIS), M5 model tree (M5 Tree), support vector regression (SVR), random forest (RF), and the powerful ensemble techniques based on Bayesian model averaging (BMA), and simple averaging (SA) to aid in improving rice yield prediction more precisely at the ripening stage. Results The fi ndings demonstrate that the ensemble model based on BMA excelled all other models on all evaluation criteria. BMA accomplished the most accurate yield prediction with correlation coeffi cient, root mean squared error (RMSE), normalized RMSE, mean absolute error, median absolute deviation, index of agreement, and a-10 values of 0.958, 0.187 t ha −1 , 0.031, 0.158 t ha −1 , 0.088 t ha −1 , 0.957, and 1.00, respectively. Conclusion Employing a combination of ML algorithms for predicting rice grain yield using UAV-based remote sensing proves to be a powerful and eff ective approach. The ensemble method improves forecast accuracy, mitigates individual algorithm limitations, and produces trustworthy outcomes for smart agricultural decisions by integrating the strengths of multiple algorithms. This comprehensive technique has the potential to adapt rice yield estimation and contribute to sustainable food production systems.

Integrating UAV Remote Sensing with GIS for Predicting Rice Grain Protein

( Tapash Kumar Sarkar ),( Chan-seok Ryu ),( Ye-seong Kang ),( Seong-heon Kim ),( Sae-rom Jeon ),( Si-hyeong Jang ),( Jun-woo Park ),( Suk-gu Kim ),( Hyun-jin Kim ) 한국농업기계학회 2018 바이오시스템공학 Vol.43 No.2

Purpose: Unmanned air vehicle (UAV) remote sensing was applied to test various vegetation indices and make prediction models of protein content of rice for monitoring grain quality and proper management practice. Methods: Image acquisition was carried out by using NIR (Green, Red, NIR), RGB and RE (Blue, Green, Red-edge) camera mounted on UAV. Sampling was done synchronously at the geo-referenced points and GPS locations were recorded. Paddy samples were air-dried to 15% moisture content, and then dehulled and milled to 92% milling yield and measured the protein content by near-infrared spectroscopy. Results: Artificial neural network showed the better performance with R² (coefficient of determination) of 0.740, NSE (Nash-Sutcliffe model efficiency coefficient) of 0.733 and RMSE (root mean square error) of 0.187% considering all 54 samples than the models developed by PR (polynomial regression), SLR (simple linear regression), and PLSR (partial least square regression). PLSR calibration models showed almost similar result with PR as 0.663 (R²) and 0.169% (RMSE) for cloud-free samples and 0.491 (R²) and 0.217% (RMSE) for cloud-shadowed samples. However, the validation models performed poorly. This study revealed that there is a highly significant correlation between NDVI (normalized difference vegetation index) and protein content in rice. For the cloud-free samples, the SLR models showed R² = 0.553 and RMSE = 0.210%, and for cloud-shadowed samples showed 0.479 as R² and 0.225% as RMSE respectively. Conclusion: There is a significant correlation between spectral bands and grain protein content. Artificial neural networks have the strong advantages to fit the nonlinear problem when a sigmoid activation function is used in the hidden layer. Quantitatively, the neural network model obtained a higher precision result with a mean absolute relative error (MARE) of 2.18% and root mean square error (RMSE) of 0.187%.

Artificial Neural Network-based Model for Predicting Moisture Content in Rice Using UAV Remote Sensing Data

( Tapash Kumar Sarkar ),( Chan-seok Ryu ),( Jeong-gyun Kang ),( Ye-seong Kang ),( Sae-rom Jun ),( Si-hyeong Jang ),( Jun-woo Park ),( Hye-young Song ) 대한원격탐사학회 2018 大韓遠隔探査學會誌 Vol.34 No.4

The percentage of moisture content in rice before harvest is crucial to reduce the economic loss in terms of yield, quality and drying cost. This paper discusses the application of artificial neural network (ANN) in developing a reliable prediction model using the low altitude fixed-wing unmanned air vehicle (UAV) based reflectance value of green, red, and NIR and statistical moisture content data. A comparison between the actual statistical data and the predicted data was performed to evaluate the performance of the model. The correlation coefficient (R) is 0.862 and the mean absolute percentage error (MAPE) is 0.914% indicate a very good accuracy of the model to predict the moisture content in rice before harvest. The model predicted values are matched well with the measured values (R² = 0.743, and Nash-Sutcliffe Efficiency = 0.730). The model results are very promising and show the reliable potential to predict moisture content with the error of prediction less than 7%. This model might be potentially helpful for the rice production system in the field of precision agriculture (PA).

Artificial Neural Network-based Model for Predicting Moisture Content in Rice Using UAV Remote Sensing Data

Sarkar, Tapash Kumar,Ryu, Chan-Seok,Kang, Jeong-Gyun,Kang, Ye-Seong,Jun, Sae-Rom,Jang, Si-Hyeong,Park, Jun-Woo,Song, Hye-Young The Korean Society of Remote Sensing 2018 大韓遠隔探査學會誌 Vol.34 No.4

The percentage of moisture content in rice before harvest is crucial to reduce the economic loss in terms of yield, quality and drying cost. This paper discusses the application of artificial neural network (ANN) in developing a reliable prediction model using the low altitude fixed-wing unmanned air vehicle (UAV) based reflectance value of green, red, and NIR and statistical moisture content data. A comparison between the actual statistical data and the predicted data was performed to evaluate the performance of the model. The correlation coefficient (R) is 0.862 and the mean absolute percentage error (MAPE) is 0.914% indicate a very good accuracy of the model to predict the moisture content in rice before harvest. The model predicted values are matched well with the measured values($R^2=0.743$, and Nash-Sutcliffe Efficiency = 0.730). The model results are very promising and show the reliable potential to predict moisture content with the error of prediction less than 7%. This model might be potentially helpful for the rice production system in the field of precision agriculture (PA).

Integrating UAV Remote Sensing with GIS for Predicting Rice Grain Protein

Sarkar, Tapash Kumar,Ryu, Chan-Seok,Kang, Ye-Seong,Kim, Seong-Heon,Jeon, Sae-Rom,Jang, Si-Hyeong,Park, Jun-Woo,Kim, Suk-Gu,Kim, Hyun-Jin Korean Society for Agricultural Machinery 2018 바이오시스템공학 Vol.43 No.2

Purpose: Unmanned air vehicle (UAV) remote sensing was applied to test various vegetation indices and make prediction models of protein content of rice for monitoring grain quality and proper management practice. Methods: Image acquisition was carried out by using NIR (Green, Red, NIR), RGB and RE (Blue, Green, Red-edge) camera mounted on UAV. Sampling was done synchronously at the geo-referenced points and GPS locations were recorded. Paddy samples were air-dried to 15% moisture content, and then dehulled and milled to 92% milling yield and measured the protein content by near-infrared spectroscopy. Results: Artificial neural network showed the better performance with $R^2$ (coefficient of determination) of 0.740, NSE (Nash-Sutcliffe model efficiency coefficient) of 0.733 and RMSE (root mean square error) of 0.187% considering all 54 samples than the models developed by PR (polynomial regression), SLR (simple linear regression), and PLSR (partial least square regression). PLSR calibration models showed almost similar result with PR as 0.663 ($R^2$) and 0.169% (RMSE) for cloud-free samples and 0.491 ($R^2$) and 0.217% (RMSE) for cloud-shadowed samples. However, the validation models performed poorly. This study revealed that there is a highly significant correlation between NDVI (normalized difference vegetation index) and protein content in rice. For the cloud-free samples, the SLR models showed $R^2=0.553$ and RMSE = 0.210%, and for cloud-shadowed samples showed 0.479 as $R^2$ and 0.225% as RMSE respectively. Conclusion: There is a significant correlation between spectral bands and grain protein content. Artificial neural networks have the strong advantages to fit the nonlinear problem when a sigmoid activation function is used in the hidden layer. Quantitatively, the neural network model obtained a higher precision result with a mean absolute relative error (MARE) of 2.18% and root mean square error (RMSE) of 0.187%.

Predicting Grain Protein Content of Rice using Remote Sensing Technology

( Tapash Kumar Sarkar ),( Chan-seok Ryu ),( Jeong-gyun Kang ),( Ye-seong Kang ),( Seon-hgeon Kim ),( Sae-rom Jun ),( Won-jun Kim ),( Suk-ku Kim ),( Hyun-jin Kim ) 한국농업기계학회 2016 한국농업기계학회 학술발표논문집 Vol.21 No.2

The ability to estimate and map grain protein content before harvest using remote sensing can help crop growers to set the strategies of harvest and management practices. The objective of the study was to make a prediction model of grain protein content of rice using remote sensing and to test various vegetation indices (normalization) on remote sensing imagery to identify the best vegetation indices for predicting the quality index. Image acquisition was carried out using NIR and RE camera mounted on drone (eBee, senseFly Ltd., Switzerland) on September 11, 2016. Sampling was done synchronously at the geo-referenced points and GPS locations were recorded. Paddy samples were air-dried to 15% moisture content then dehulled and measured the protein and amylose content of the brown rice by grain analyzer (FOSS Infratech^TM 1241 Analyzer, Hoganas, Sweden). Hereafter, each sample was milled to 92% milling yield by a polishing machine and measured the protein and amylose content again. Total images were differentiated into two groups as cloud free and cloud shadowed. On one hand for the cloud free samples; the vegetation index, NDVI derived from the canopy spectral reflectance at green, red and NIR bands, was significantly correlated to the final grain protein content (R2=0.553, RMSE=0.21, n=14) at 99% significant level. On the other hand, for cloud shadowed samples, the result demonstrated that vegetation index, NDVI was significantly correlated to the final grain protein content(R2=0.479, RMSE=0.225, n=35) at 99.9% significant level. Grain protein content of rice can be possibly forecasted using the canopy or images spectral reflectance of Drone at grain filling stage. These models might not be able to directly apply to different ecological conditions, because this study was carried out under the same ecological conditions in the same district. However, use of remote sensing data to predict quality indices such as protein content of grains can be feasible and realized.

Estimating Moisture Content of Cucumber Seedling Using Hyperspectral Imagery

강정균,류찬석,김성훈,강예성,Tapash Kumar Sarkar,강동현,김동억,구양규 한국농업기계학회 2016 바이오시스템공학 Vol.41 No.3

Purpose: This experiment was conducted to detect water stress in terms of the moisture content of cucumber seedlings under water stress condition using a hyperspectral image acquisition system, linear regression analysis, and partial least square regression (PLSR) to achieve a non-destructive measurement procedure. Methods: Changes in the reflectance spectrum of cucumber seedlings under water stress were measured using hyperspectral imaging techniques. A model for estimating moisture content of cucumber seedlings was constructed through a linear regression analysis that used the moisture content of cucumber seedlings and a normalized difference vegetation index (NDVI). A model using PLSR that used the moisture content of cucumber seedlings and reflectance spectrum was also created. Results: In the early stages of water stress, cucumber seedlings recovered completely when sub-irrigation was applied. However, the seedlings suffering from initial wilting did not recover when more than 42 h passed without irrigation. The reflectance spectrum of seedlings under water stress decreased gradually, but increased when irrigation was provided, except for the seedlings that had permanently wilted. From the results of the linear regression analysis using the NDVI, the model excluding wilted seedlings with less than 20% (n=97) moisture content showed a precision (R² and R²a) of 0.573 and 0.568, respectively, and accuracy (RE) of 4.138% and 4.138%, which was higher than that for models including all seedlings (n=100). For PLS regression analysis using the reflectance spectrum, both models were found to have strong precision (R²) with a rating of 0.822, but accuracy (RMSE and RE) was higher in the model excluding wilted seedlings as 5.544% and 13.65% respectively. Conclusions: The estimation model of the moisture content of cucumber seedlings showed better results in the PLSR analysis using reflectance spectrum than the linear regression analysis using NDVI.

Estimation of Leaf Dry Mass and Nitrogen Content for Soybean using Multi-spectral Camera Mounted on Unmanned Aerial Vehicle

강예성,김성헌,강정균,홍영기,Tapash Kumar Sarkar,류찬석 경상대학교 농업생명과학연구원 2016 농업생명과학연구 Vol.50 No.6

Recently, remote sensing technology as a nondestructive method has been utilized to detectthe quantity and quality of crops using unmanned aerial system. To predict vegetation growth(leaf dry mass and nitrogen content) of soybean, two vegetation index(NDVI and Green NDVI)were calculated from images acquired by multi-spectral camera mounted on a UAV and eachprediction models between vegetation growth and index were evaluated. As a result, there wasno significant difference between vegetation growth and index when each vegetation stage foreach yellow and black bean were compared to each other. However, there was significantdifference between vegetation growth and index when all vegetation stage for each yellow andblack bean were compared to each other. Moreover, there was significant difference betweenvegetation growth and NDVI(r= 0.799 for leaf dry mass, r= 0.796 for nitrogen content), andGreen NDVI(r= 0.860 for leaf dry mass, r= 0.845 for nitrogen content) for all vegetation stageswith all soybeans. The accuracy and precision of Green NDVI model(R2= 0.740 for leaf drymass, R2= 0.714 for nitrogen content) were better than those of NDVI model regardless ofvarieties and vegetation growth. Therefore, Green NDVI has considerable potential to detect thequantity and quality of soybeans.

Yield Prediction of Chinese Cabbage (Brassicaceae) Using Broadband Multispectral Imagery Mounted Unmanned Aerial System in the Air and Narrowband Hyperspectral Imagery on the Ground

강예성,류찬석,김성헌,전새롬,장시형,박준우,Tapash Kumar Sarkar,송혜영 한국농업기계학회 2018 바이오시스템공학 Vol.43 No.2

Purpose: A narrowband hyperspectral imaging sensor of high-dimensional spectral bands is advantageous for identifying the reflectance by selecting the significant spectral bands for predicting crop yield over the broadband multispectral imaging sensor for each wavelength range of the crop canopy. The images acquired by each imaging sensor were used to develop the models for predicting the Chinese cabbage yield. Methods: The models for predicting the Chinese cabbage (Brassica campestris L.) yield, with multispectral images based on unmanned aerial vehicle (UAV), were developed by simple linear regression (SLR) using vegetation indices, and forward stepwise multiple linear regression (MLR) using four spectral bands. The model with hyperspectral images based on the ground were developed using forward stepwise MLR from the significant spectral bands selected by dimension reduction methods based on a partial least squares regression (PLSR) model of high precision and accuracy. Results: The SLR model by the multispectral image cannot predict the yield well because of its low sensitivity in high fresh weight. Despite improved sensitivity in high fresh weight of the MLR model, its precision and accuracy was unsuitable for predicting the yield as its R2 is 0.697, root-mean-square error (RMSE) is 1170 g/plant, relative error (RE) is 67.1%. When selecting the significant spectral bands for predicting the yield using hyperspectral images, the MLR model using four spectral bands show high precision and accuracy, with 0.891 for R2, 616 g/plant for the RMSE, and 35.3% for the RE. Conclusions: Little difference was observed in the precision and accuracy of the PLSR model of 0.896 for R2, 576.7 g/plant for the RMSE, and 33.1% for the RE, compared with the MLR model. If the multispectral imaging sensor composed of the significant spectral bands is produced, the crop yield of a wide area can be predicted using a UAV.

Estimating Moisture Content of Cucumber Seedling Using Hyperspectral Imagery

Kang, Jeong-Gyun,Ryu, Chan-Seok,Kim, Seong-Heon,Kang, Ye-Seong,Sarkar, Tapash Kumar,Kang, Dong-Hyeon,Kim, Dong Eok,Ku, Yang-Gyu Korean Society for Agricultural Machinery 2016 바이오시스템공학 Vol.41 No.3

Purpose: This experiment was conducted to detect water stress in terms of the moisture content of cucumber seedlings under water stress condition using a hyperspectral image acquisition system, linear regression analysis, and partial least square regression (PLSR) to achieve a non-destructive measurement procedure. Methods: Changes in the reflectance spectrum of cucumber seedlings under water stress were measured using hyperspectral imaging techniques. A model for estimating moisture content of cucumber seedlings was constructed through a linear regression analysis that used the moisture content of cucumber seedlings and a normalized difference vegetation index (NDVI). A model using PLSR that used the moisture content of cucumber seedlings and reflectance spectrum was also created. Results: In the early stages of water stress, cucumber seedlings recovered completely when sub-irrigation was applied. However, the seedlings suffering from initial wilting did not recover when more than 42 h passed without irrigation. The reflectance spectrum of seedlings under water stress decreased gradually, but increased when irrigation was provided, except for the seedlings that had permanently wilted. From the results of the linear regression analysis using the NDVI, the model excluding wilted seedlings with less than 20% (n=97) moisture content showed a precision ($R^2$ and $R^2_{\alpha}$) of 0.573 and 0.568, respectively, and accuracy (RE) of 4.138% and 4.138%, which was higher than that for models including all seedlings (n=100). For PLS regression analysis using the reflectance spectrum, both models were found to have strong precision ($R^2$) with a rating of 0.822, but accuracy (RMSE and RE) was higher in the model excluding wilted seedlings as 5.544% and 13.65% respectively. Conclusions: The estimation model of the moisture content of cucumber seedlings showed better results in the PLSR analysis using reflectance spectrum than the linear regression analysis using NDVI.