AI for Rice Farming: Seedling Growth at Nursery

Artificial Intelligence technology offers significant advancements in the growth of rice seedlings at nurseries, optimizing various aspects of cultivation to enhance efficiency, quality, and yield.

Implementing AI for rice seedling growth at nurseries is important. For example, such if farmer use drones and smartphone to monitor seedling health and growth stages, it enhances precision in optimizing growing conditions and build the first yield forecast for the season.

This is why today we concentrate our attention on such an important topic as monitoring rice seedling growth.

Rice Growth Stage Classification with Machine Learning & Image Processing

The first study for today has been published by the research team from Taiwan ????. They did an attempt to classify rice growth stages using random forest-based machine learning and image processing techniques to support precision agriculture.

The research employed high-resolution cameras and IoT devices to capture images and environmental data from rice fields. Specifically, the researchers used:

• HD smart cameras (Speed-dome)
• High-resolution RGB cameras
• 7-in-1 soil sensors
• Weather monitoring stations
• Flow meters
• Milometers
• LoRa base stations

Using the data collected with these devices, various image processing techniques like object detection, instance segmentation, and green index calculation were applied to determine growth parameters such as canopy cover and plant height. These parameters, along with environmental data, were fed into a random forest (RF) machine learning model to classify the growth stages of rice.

The RF-based model achieved high performance with an accuracy of 99.45% and a macro F1-score of 97.34%.

Key features contributing to this performance included accumulated temperature, days after transplanting (DAT), and plant height.

The model's robustness was further improved using up-sampling methods like SMOTE-ENN to address data imbalances, resulting in an F1-score of 98.65%. This approach significantly enhanced the ability to accurately classify various growth stages, demonstrating the potential of integrating image processing with machine learning for precision agriculture.

These findings can be practically applied by agronomists, rice farmers, and agricultural researchers to improve rice crop management, optimize resource use, and enhance yield prediction through precise growth stage monitoring.

Machine Learning Approaches for Rice Seedling Growth Stages Detection

This study, published by Chinese ???? research team, aimed to develop machine learning models for detecting the growth stages of rice seedlings using UAV images to improve efficiency and accuracy over manual methods.

The research involved capturing UAV images of rice seedlings at three critical growth stages (BBCH11, BBCH12, and BBCH13) and employing both traditional machine learning (HOG-SVM) and deep learning algorithms (EfficientNet and other CNNs). The dataset included images of three rice cultivars, five nursing densities, and various sowing dates.

Key techniques, deployed by the researchers, included histograms of oriented gradients (HOG) for feature extraction and support vector machines (SVM) for classification, alongside deep learning models like EfficientNetB4, VGG16, ResNet50, and DenseNet121.

The study demonstrated that deep learning models significantly outperformed traditional HOG-SVM methods in classifying rice seedling growth stages. The EfficientNetB4 model achieved the highest accuracy, average precision, average recall, and F1 score, with values of 99.47%, 99.53%, 99.39%, and 99.46%, respectively. In contrast, the best HOG-SVM model reached accuracy, average precision, average recall, and F1 score values of 84.9%, 85.9%, 84.9%, and 85.4%, respectively. These results indicate that integrating UAV imagery with advanced machine learning algorithms can provide a robust and efficient tool for monitoring rice seedling development.

These findings are particularly useful for agricultural researchers, agronomists, and rice producers aiming to optimize rice nursery management and improve crop yields through precise and automated growth stage detection. For example, this method can be adapted for monitoring rice seedlings with a smartphone by leveraging the device's camera to capture images and utilizing mobile machine learning applications to analyze and classify the growth stages based on the collected data

Estimation of Rice Seedling Growth Traits with Deep Learning

The final study, considered today was conducted by the Chinese ???? researchers, to accurately assess the growth-related traits of rice seedlings in a controlled environment using an end-to-end multi-objective deep learning framework.

The researchers used convolutional neural networks (CNNs) to analyze RGB images of rice seedlings, collected with a Nikon Z5 camera and an iPhone 12. The images were processed through a two-stage model: a U-Net for segmentation and a modified ResNet50 for regression to predict shoot height (SH) and shoot fresh weight (SFW).

The model was trained using a dataset of 984 images, with data augmentation and K-fold cross-validation to ensure robustness.

The proposed CNN model achieved high accuracy in estimating growth traits, with R2 values of 0.980 for shoot height and 0.717 for fresh weight. The normalized root mean square error (NRMSE) was 2.64% for height and 17.23% for weight, outperforming traditional methods like random forests and simpler regression CNN models. These results indicate that the model can effectively predict rice seedling growth traits from digital images.

Conclusions

• Implementing AI technology, such as machine learning models and image processing techniques, significantly improves the precision and efficiency of monitoring rice seedling growth stages, enabling optimized growing conditions and accurate yield predictions. For instance, a Chinese research team developed deep learning models that achieved an accuracy of 99.47% in classifying growth stages using UAV images, with corresponding precision, recall, and F1 scores of 99.53%, 99.39%, and 99.46%, respectively, significantly outperforming traditional HOG-SVM methods which had accuracy values of 84.9%.

• The use of high-resolution cameras, IoT devices, and environmental sensors in conjunction with AI provides a robust framework for monitoring rice seedling growth. A study by a Taiwanese research team utilized HD smart cameras, RGB cameras, 7-in-1 soil sensors, weather monitoring stations, flow meters, milometers, and LoRa base stations to collect detailed data on rice seedlings. These data were analyzed using random forest-based machine learning models, achieving a high performance with an accuracy of 99.45% and a macro F1-score of 97.34%, further improved to 98.65% with up-sampling methods like SMOTE-ENN.

• These AI-driven methods can be practically applied by agricultural researchers, agronomists, and rice farmers to enhance nursery management and crop yields. For example, smartphone integration can leverage camera capabilities to capture images and use mobile machine learning applications to analyze growth stages. A study showed that using a CNN model to estimate growth traits from digital images collected with a Nikon Z5 camera and an iPhone 12 achieved R2 values of 0.980 for shoot height and 0.717 for shoot fresh weight, indicating high accuracy and practical utility in real-world applications.

?? What's next in AI for Rice Farming?

In the next 'AI for Rice Farming' edition, we will explore upland rice. We'll try to find some interesting research papers that demonstrate the successful application of innovative technologies for rice which is cultivated in rainfed, naturally well-drained fields, within mixed farming systems that do not involve irrigation or puddling.

?What do you think about this topic?

Please, share your thoughts with us in the comments below ??

Wishes of high germination rate and good density of your rice seedlings,

Maryna Kuzmenko, Ph.D ????, Chief Inspiration Officer at Petiole Pro Community

#rice

Photo credit for the cover:

Sheng, R.T.-C.; Huang, Y.-H.; Chan, P.-C.; Bhat, S.A.; Wu, Y.-C.; Huang, N.-F. Rice Growth Stage Classification via RF-Based Machine Learning and Image Processing. Agriculture 2022, 12, 2137. https://doi.org/10.3390/agriculture12122137