Deep Learning for Classification and Severity Estimation of Coffee Leaf Biotic Stress (1907.11561v1)

Abstract: Biotic stress consists of damage to plants through other living organisms. Efficient control of biotic agents such as pests and pathogens (viruses, fungi, bacteria, etc.) is closely related to the concept of agricultural sustainability. Agricultural sustainability promotes the development of new technologies that allow the reduction of environmental impacts, greater accessibility to farmers and, consequently, increase on productivity. The use of computer vision with deep learning methods allows the early and correct identification of the stress-causing agent. So, corrective measures can be applied as soon as possible to mitigate the problem. The objective of this work is to design an effective and practical system capable of identifying and estimating the stress severity caused by biotic agents on coffee leaves. The proposed approach consists of a multi-task system based on convolutional neural networks. In addition, we have explored the use of data augmentation techniques to make the system more robust and accurate. The experimental results obtained for classification as well as for severity estimation indicate that the proposed system might be a suitable tool to assist both experts and farmers in the identification and quantification of biotic stresses in coffee plantations.

• The paper introduces a multi-task deep learning model using CNNs to classify coffee leaf biotic stress and estimate its severity.
• It applies advanced data augmentation techniques, including mixup, to enhance model robustness on a specialized dataset of 1747 images.
• ResNet50 outperforms other architectures by achieving up to 96.63% accuracy on isolated symptoms, underscoring the impact of high-resolution data.

Deep Learning for Classification and Severity Estimation of Coffee Leaf Biotic Stress

The paper "Deep Learning for Classification and Severity Estimation of Coffee Leaf Biotic Stress" presents a thorough investigation into the utilization of deep learning methodologies, specifically convolutional neural networks (CNNs), to address the challenge of identifying and assessing the severity of biotic stresses on coffee leaves. Biotic stresses, including pests and pathogens like fungi and bacteria, can lead to substantial crop losses. The authors emphasize the necessity of early and precise identification of these stress factors to mitigate their impact efficiently, aligning with the broader goal of agricultural sustainability.

Methodology and Experimental Design

The research introduces a multi-task learning system leveraging CNNs as its core. This system aims to concurrently perform two tasks: (1) diagnoses the type of biotic stress and (2) estimates the severity of this stress. The use of multi-task learning is strategically chosen to exploit shared representations between related tasks, potentially improving the model's generalization and reducing overfitting.

Data augmentation techniques have been implemented to enhance model robustness, crucial given the relatively small dataset size. Among these techniques, the standard augmentation (mirror, rotation, color variations) and mixup, which involves image synthesis through linear combination, are highlighted.

The research experiment encompasses various well-established CNN architectures, including AlexNet, GoogLeNet, VGG19, and ResNet50, focusing on balancing model complexity and performance.

Dataset and Implementation

The authors have created a specialized dataset for this paper, capturing 1747 images of coffee leaves exhibiting key biotic stresses: leaf miner, rust, brown leaf spot, and cercospora leaf spot. The data stems from photographs taken under partially controlled conditions using various smartphone models. A dual dataset approach is undertaken: a leaf dataset, representing the full leaf images, and a symptom dataset, isolating individual stress symptoms to assist in detailed classification tasks.

Results and Discussion

The ResNet50 architecture with multi-task learning demonstrated superior performance, achieving an accuracy of 94.05% for biotic stress classification and 84.76% for severity estimation when using leaf datasets. Interestingly, the classification accuracy reached 96.63% when symptoms were isolated. These results underscore the importance of focused and high-resolution symptom data in achieving accurate diagnostics.

The paper also identifies cercospora leaf spot as a particularly challenging biotic stress to classify, likely due to its visual similarity to other conditions and the dataset's class imbalance.

Implications and Future Directions

The findings present a practical framework for assisting farmers and experts in diagnosing and managing coffee diseases efficiently. While the current dataset covers major stress factors, expanding it to incorporate a broader spectrum of stress types and increasing sample diversity remains critical. Additionally, real-world application through smartphone integration is ongoing, reflecting the potential of this technology to transform agricultural practices.

The exploration of multi-task learning in this context offers promising avenues for further research, particularly in expanding its applicability to other crops and stress types. The integration of advanced data augmentation techniques like mixup shows potential in addressing dataset limitations, though its impact on severity estimation requires further scrutiny.

In summary, this research contributes significantly to applying deep learning methods in agriculture, providing a robust baseline for future efforts aiming to integrate AI solutions in sustainable farming practices. The continuous refinement and field implementation of this system could substantially enhance crop resilience and productivity.