Related Experiment Videos
Reducing catastrophic forgetting in CNNs for plant stress classification using continual learning
Ahmed Shakib Reza1, Shakib Sadat Shanto2, Farah Ulfat Zaima1
1Department of Computer Science and Engineering, BRAC University (BRACU), Dhaka, 1212, Bangladesh.
Scientific Reports
|June 27, 2026
Summary
Lightweight Convolutional Neural Networks (CNNs) with continual learning (CL) methods like EWC and LwF effectively classify plant stresses. This approach overcomes catastrophic forgetting in AI models for agriculture, even with limited resources.
Area of Science:
- Agricultural AI
- Computer Vision
- Machine Learning
Background:
- Manual plant stress detection is slow and error-prone, impacting crop yields, especially in regions like Bangladesh.
- Convolutional Neural Networks (CNNs) show promise for plant leaf classification but struggle with catastrophic forgetting in sequential learning scenarios.
- Existing continual learning (CL) studies often use heavy models, posing challenges for resource-constrained agricultural applications.
Purpose of the Study:
- To investigate the effectiveness of lightweight CNNs integrated with CL techniques for plant stress classification.
- To evaluate if established CL methods can mitigate catastrophic forgetting in resource-limited AI systems for agriculture.
- To assess the performance of Elastic Weight Consolidation (EWC) and Learning without Forgetting (LwF) using the EfficientNet-B0 backbone.
Main Methods:
- Integrated Elastic Weight Consolidation (EWC) and Learning without Forgetting (LwF) into the EfficientNet-B0 lightweight CNN architecture.
- Evaluated the system on the Nutrispace cucurbit nutritional deficiency dataset with three sequential tasks (ash gourd, bitter gourd, snake gourd).
- Each task involved classifying three conditions: healthy, nitrogen deficiency, and potassium deficiency.
Main Results:
- Without CL, accuracy on the initial task dropped to 30%.
- EWC maintained over 61% accuracy on prior tasks while achieving 98% on the final task.
- LwF achieved 98% on the final task, with slightly reduced retention on earlier tasks. Both methods significantly outperformed the baseline.
Conclusions:
- Lightweight CNNs combined with CL techniques (EWC, LwF) provide a viable solution for plant stress classification under resource constraints.
- EWC demonstrated superior retention of prior-task knowledge compared to LwF.
- This research offers a practical pathway for developing efficient AI tools for precision agriculture.
Related Concept Videos
01:57Adaptations that Reduce Water Loss
Though evaporation from plant leaves drives transpiration, it also results in loss of water. Because water is critical for photosynthetic reactions and other cellular processes, evolutionary pressures on plants in different environments have driven the acquisition of adaptations that reduce water loss.
01:19Survival Tree
Survival trees are a non-parametric method used in survival analysis to model the relationship between a set of covariates and the time until an event of interest occurs, often referred to as the "time-to-event" or "survival time." This method is particularly useful when dealing with censored data, where the event has not occurred for some individuals by the end of the study period, or when the exact time of the event is unknown.
Building a Survival Tree
Constructing a survival tree begins...
Building a Survival Tree
Constructing a survival tree begins...
01:27C4 Pathway and CAM
Most plants use the C3 pathway for carbon fixation. However, some plants, such as sugar cane, corn, and cacti that grow in hot conditions, use alternative pathways to fix carbon and conserve energy loss due to photorespiration. Photorespiration is the process that occurs when the oxygen concentration is high. Under such conditions, the rubisco enzyme in the Calvin cycle binds O2 instead of CO2, which halts photosynthesis and consumes energy.
C4 Pathway
The C4 pathway is used by plants such as...
C4 Pathway
The C4 pathway is used by plants such as...
01:21Cognitive Learning
Cognitive learning is based on purposive behavior, incidental learning, and insight learning.
E. C. Tolman's theory of purposive behavior emphasizes that much behavior is goal-directed. He argued that to understand behavior, we must look at the entire sequence of actions leading to a goal. For instance, high school students study hard, not just due to past reinforcement but also to achieve the goal of getting into a good college.
Tolman introduced the idea that behavior is influenced by...
E. C. Tolman's theory of purposive behavior emphasizes that much behavior is goal-directed. He argued that to understand behavior, we must look at the entire sequence of actions leading to a goal. For instance, high school students study hard, not just due to past reinforcement but also to achieve the goal of getting into a good college.
Tolman introduced the idea that behavior is influenced by...
02:41Responses to Drought and Flooding
Water plays a significant role in the life cycle of plants. However, insufficient or excess of water can be detrimental and pose a serious threat to plants.
02:45Responses to Heat and Cold Stress
Every organism has an optimum temperature range within which healthy growth and physiological functioning can occur. At the ends of this range, there will be a minimum and maximum temperature that interrupt biological processes.