??AI for Okra: Water Stress Identification ??

If you'd like to receive the regular 'AI in Agriculture' newsletter in your inbox, simply add your email to my mailing list.

Join over 8,742 readers who enjoy weekly updates on AI advancements in agriculture!

To get more information about okra phenotyping capabilities with mobile devices visit Petiole Pro

Why Okra?

Okra (Abelmoschus esculentus L.) is a globally significant vegetable crop, with annual production of over 10.5 million tons, of which 60% is cultivated in India, making it an economically vital crop, especially in the country’s coastal and central regions.

Its high sensitivity to water stress significantly impacts yield, making it an ideal candidate for research focused on irrigation optimization and water stress management.

Global production of Okra

The top five producers of okra worldwide are:

1. India ???? – The largest producer, contributing over 60% of global okra production.
2. Nigeria ???? – A major producer in Africa, with a significant portion of its agricultural output dedicated to okra.
3. Sudan ???? – A key okra producer in Northeast Africa, supplying both domestic and regional markets.
4. C?te d'Ivoire ???? – A leading producer in West Africa, with okra being a staple in local cuisine.
5. Pakistan ???? – A prominent producer in South Asia, where okra is an essential component of the diet and agriculture.

In the figure above: (a) NGB00386; (b) NGB00371; (c) NGB00486; (d) NGB00345; (e) NGB00470; (f) NGB00378A; (g) NGB00331; (h) NGB00356; (i) NGB00430; (j) NGB00299; (k) NGB00355; (l) NGB00304; (m) NGB00308.

The main challenges in water stress identification in Okra and similar crops

1. Complexity of Plant Physiological Response

Crops exhibit a range of physiological responses to water stress, including changes in leaf colour, structure, and temperature.

These responses vary based on the plant’s developmental stage and environmental factors, making it difficult to develop universal indicators for early detection and classification of water stress.

In the figure above: histochemical localization of oxidative stress markers H2O2 and O2?, as affected by drought stress at 5 and 10 days of interval in okra genotypes NS7774, NS7772, Green Gold, and OH3312, along with respective controls. The brownish colour on the leaves indicates the localization of H2O2 stress marker, whereas the bluish colour indicates the localization of O2?1 marker.

In the figure above: Vascular transport, as affected by drought stress at 5 and 10 days of interval in okra genotypes NS7774, NS7772, Green Gold, and OH3312, along with respective controls. Red and blue food-coloring dyes were used for the absorption process, indicating activity of vascular tissue. EP indicates epidermis; CR indicates cortex; X indicates xylem; P indicates phloem.

In the figure above: Representative images of stomata, as affected by drought stress at 5 and 10 days of interval in okra genotypes NS7774, NS7772, Green Gold, and OH3312, along with respective controls observed under scanning electron microscope (SEM) at 40X magnification. In images, GC indicates guard cells and SP indicates stomatal pore.

2. Limitations of Imaging Technologies

While thermal and RGB imaging are widely used for stress detection, they have limitations. RGB imaging is sensitive to lighting conditions and canopy structure, whereas thermal imaging can be influenced by ambient temperature and humidity. These factors can lead to inconsistencies in data, making accurate water stress assessment challenging. Thermal imagery has been successfully applied on other stages of okra production.

3. Scalability and Resource Constraints

Implementing water stress detection at scale requires a significant investment in sensors, imaging systems, and computational resources.

Additionally, developing models that work effectively across different crop types, soil conditions, and geographical regions remains challenging due to the diversity in agricultural environments and limited field data availability.

Thermal–RGB Imagery and Computer Vision for Water Stress Identification of Okra (Abelmoschus esculentus L.)

Country: India ????, United States ????

Published: 27 June 2024

This study aims to use thermal and RGB imagery, combined with deep learning models, to identify water stress in okra crops and develop precision irrigation management systems. The researchers explore the performance of ResNet-50 and MobileNetV2 deep learning models on 3,200 images to detect stress levels under varying irrigation treatments.

The primary question researchers sought to answer was: Can thermal and RGB imagery integrated with deep learning techniques accurately identify water stress in okra crops under different irrigation treatments?        

For the experiment, a two-year field study was conducted with four irrigation levels (100%, 75%, 50%, and 25% of crop evapotranspiration) and two irrigation methods (flood and sprinkler).

Thermal and RGB images of okra plants were captured using a Krykard TCA 1950 thermal camera and a Canon EOS 3000D camera at different crop stages. These images were processed using ResNet-50 and MobileNetV2, two deep learning models. Each image dataset was split into training, validation, and testing subsets, and the models' performance was evaluated based on precision, sensitivity, and F1 score metrics.

Application of AI for Okra water stress assessment

AI was at the core of this research. The use of convolutional neural networks (CNNs) allowed for automated feature extraction from the images, distinguishing water-stressed from non-stressed crops.

Two models, ResNet-50 and MobileNetV2, were trained separately on RGB and thermal images:

1. ResNet-50 showed superior accuracy in thermal imagery classification.
2. MobileNetV2 was more computationally efficient, suitable for deployment on mobile devices.

In figure above: Relative water content (a), canopy temperature (b), soil moisture content (c), and relative humidity (d) under different irrigation levels observed during okra growing season over two years. F-Flood and S-sprinkler irrigation method at 100, 75, 50, and 25% crop evapotranspiration.

Key findings of the research paper

• Thermal imagery resulted in higher water stress identification accuracy (87.9% with ResNet-50, 84.3% with MobileNetV2) compared to RGB imagery (78.6% and 74.1%, respectively).
• Maximum okra yield was 10,666 kg/ha for flood irrigation and 9876 kg/ha for sprinkler irrigation at 100% irrigation level, with crop water use efficiency of 1.16 kg/m3 and 1.24 kg/m3, respectively.
• These results highlight the potential for AI-driven irrigation systems to optimize water use and crop yield.

Agricultural researchers and precision irrigation practitioners can practically apply these findings. The insights into water stress detection can guide the development of automated irrigation systems that improve water use efficiency.

Technologies used

???? Hardware:

1. Krykard TCA 1950 Thermal Camera
2. Canon EOS 3000D Camera
3. Micro-sprinklers (Netafim, India)

???? Software and Models:

1. ResNet-50
2. MobileNetV2
3. Convolutional Neural Networks (CNNs)

References in today's edition of "AI for Okra: Water Stress Identification"

FYI (For Your Interest)