CD&S Dataset: Handheld Imagery Dataset Acquired Under Field Conditions for Corn Disease Identification and Severity Estimation (2110.12084v1)

Abstract: Accurate disease identification and its severity estimation is an important consideration for disease management. Deep learning-based solutions for disease management using imagery datasets are being increasingly explored by the research community. However, most reported studies have relied on imagery datasets that were acquired under controlled lab conditions. As a result, such models lacked the ability to identify diseases in the field. Therefore, to train a robust deep learning model for field use, an imagery dataset was created using raw images acquired under field conditions using a handheld sensor and augmented images with varying backgrounds. The Corn Disease and Severity (CD&S) dataset consisted of 511, 524, and 562, field acquired raw images, corresponding to three common foliar corn diseases, namely Northern Leaf Blight (NLB), Gray Leaf Spot (GLS), and Northern Leaf Spot (NLS), respectively. For training disease identification models, half of the imagery data for each disease was annotated using bounding boxes and also used to generate 2343 additional images through augmentation using three different backgrounds. For severity estimation, an additional 515 raw images for NLS were acquired and categorized into severity classes ranging from 1 (resistant) to 5 (susceptible). Overall, the CD&S dataset consisted of 4455 total images comprising of 2112 field images and 2343 augmented images.

• The paper presents the creation of a comprehensive dataset with 4455 images and field-realistic augmentation, advancing deep learning for corn disease diagnosis.
• It details precise data collection using high-resolution imagery and YOLO-compatible annotations to accurately map disease lesions.
• The research demonstrates promising applications in automated severity estimation and supports transfer learning for broader agricultural challenges.

Overview of the CD{content}S Dataset for Corn Disease Identification and Severity Estimation

The research presented in this paper addresses a significant gap in the field of automated agricultural disease management by developing the Corn Disease and Severity (CD{content}S) dataset. This dataset explicitly aims to enhance the accuracy of disease identification and severity estimation models, which is crucial for managing foliar diseases impacting corn yield. Traditional methods of disease identification heavily rely on manual scouting and expert knowledge, which are inherently time-consuming and subjective. By leveraging deep learning methodologies, the authors propose a dataset that better represents the challenging conditions found in field environments.

Contributions and Data Composition

The paper describes the creation of a comprehensive dataset comprised of 4455 images, of which 2112 are raw field images capturing symptoms of three prevalent foliar corn diseases: Northern Leaf Blight (NLB), Gray Leaf Spot (GLS), and Northern Leaf Spot (NLS). The dataset is expanded through augmentation techniques, adding 2343 images with varied background settings to improve model generalization capabilities.

Key elements of the dataset include:

• Raw Field Images: Acquired under realistic field conditions using a 12-megapixel camera, capturing images at 3000x3000 pixels. This element addresses the variability and complexity seen in natural environments compared to controlled lab settings.
• Augmented Images: Utilizes three different background configurations - black, white, and transparent - to enhance model robustness against background variation in field conditions.
• Severity Classification: For the NLS disease, an additional 515 images were collected and categorized according to a severity scale from 1 (resistant) to 5 (susceptible).

Methodological Approach

The development of this dataset involved meticulous processes for data collection, annotation, and augmentation. Images were annotated with bounding boxes to facilitate object detection tasks essential for mapping disease lesions. The choice of the YOLO annotation format ensures compatibility with state-of-the-art deep learning frameworks for both classification and detection tasks.

The augmentation strategy adopted introduces diversity into the training datasets, thus tackling the limitations imposed by small datasets and improving the predictive performance of models when deployed in variable real-world field conditions.

Implications and Future Prospects

The CD{content}S dataset provides an invaluable resource for advancing the accuracy of machine learning-based agricultural disease diagnosis systems. By bridging the gap between controlled environment datasets and field-acquired datasets, this research advances the readiness of AI systems for deployment in practical agricultural settings.

The dataset's design, which combines field realism with a controlled approach to dataset augmentation, has strong potential for transfer learning applications in related agricultural tasks. Furthermore, as the dataset specifically targets corn diseases but follows a systematic approach, it sets a precedent for creating similar datasets for other critical crops, potentially impacting global food security.

In future work, integrating this dataset with temporal data and employing sequential models could enhance severity prediction accuracy further. Moreover, exploring multispectral imaging alongside traditional RGB imagery might present new avenues to improve distinctions between various stages of disease progression, exploiting variances in reflectance properties unobservable in standard imagery.

The CD{content}S dataset represents a step forward in harmonizing research paradigms with practical agricultural applications, facilitating the development of robust AI models crucial for sustainable crop management and protection.