Arable - NemaRecognition: An AI-and molecular-driven pipeline for throughput plant parasitic nematode recognition
Lead Research Organisation:
University of Sheffield
Department Name: Computer Science
Abstract
NemaRecognition will be a machine learning based automatic image recognition technique capable of real-time detection of PPNs using digital images/videos.
Plant clinics carry out a suite of services for growers and their advisers. A key service is the assessment of soil samples for PPN. PPN screening is carried out through time-intensive taxonomic identification, this is reliant on taxonomic expertise and several years of training. Trained nematologists are in short supply, causing concern in the industry as accurate and reliable identification of PPN is a critical factor influencing agronomic decisions.
PPN affect various crops and can devastate yields, with losses up to 35% (AHDB, 2017). Growers screen fields prior to planting to identify and quantify PPN to help decide on the crop to be planted/avoided, guide variety choice, and advise control strategies. PPN screening can cost £70 per field per season and represents a substantial cost. More rapid, cost-effective assessment methods would represent a cost saving to growers.
Alternatives, such as molecular-based tests, have been developed but have substantial shortcomings in accuracy, breadth of use, and grower-confidence. AI algorithms have been developed for nematode identification; however, the majority only identify one PPN genera (Bogale et al., 2020; Akintayo et al., 2018). NemaRecognition would represent an innovative state-of-the-art solution for PPN assessment by providing recognition for multiple PPN genera, and through further development would become one of the first machine learning-based techniques providing plant health services to UK growers.
Image-recognition techniques have been developed for other agricultural pests (e.g., insects). However, significant challenges to producing a transformative PPN recognition system using machine learning techniques remain, including recognition of a range of PPN genera, detection in field samples, recognition through video-capture, validation, benchmarking, and selection of appropriate models.
NemaRecognition would have myriad benefits, including reduced grower costs (passed down through plant clinic cost savings), increased accessibility to PPN screening in regions where services are inhibited by a taxonomic skills shortage, and as a training tool to help address the taxonomic skills shortage within the industry. Global challenges have been influential in creating this opportunity: UK net-zero farming, EU Sustainable Use Directive, UK path to sustainable farming.
The NemaRecognition project will showcase the feasibility and applicability of this technology toward PPN detection and would also represent proof-of-concept for developing similar innovations for other soil-dwelling organisms, with significant potential in the growing area of soil health services.
Plant clinics carry out a suite of services for growers and their advisers. A key service is the assessment of soil samples for PPN. PPN screening is carried out through time-intensive taxonomic identification, this is reliant on taxonomic expertise and several years of training. Trained nematologists are in short supply, causing concern in the industry as accurate and reliable identification of PPN is a critical factor influencing agronomic decisions.
PPN affect various crops and can devastate yields, with losses up to 35% (AHDB, 2017). Growers screen fields prior to planting to identify and quantify PPN to help decide on the crop to be planted/avoided, guide variety choice, and advise control strategies. PPN screening can cost £70 per field per season and represents a substantial cost. More rapid, cost-effective assessment methods would represent a cost saving to growers.
Alternatives, such as molecular-based tests, have been developed but have substantial shortcomings in accuracy, breadth of use, and grower-confidence. AI algorithms have been developed for nematode identification; however, the majority only identify one PPN genera (Bogale et al., 2020; Akintayo et al., 2018). NemaRecognition would represent an innovative state-of-the-art solution for PPN assessment by providing recognition for multiple PPN genera, and through further development would become one of the first machine learning-based techniques providing plant health services to UK growers.
Image-recognition techniques have been developed for other agricultural pests (e.g., insects). However, significant challenges to producing a transformative PPN recognition system using machine learning techniques remain, including recognition of a range of PPN genera, detection in field samples, recognition through video-capture, validation, benchmarking, and selection of appropriate models.
NemaRecognition would have myriad benefits, including reduced grower costs (passed down through plant clinic cost savings), increased accessibility to PPN screening in regions where services are inhibited by a taxonomic skills shortage, and as a training tool to help address the taxonomic skills shortage within the industry. Global challenges have been influential in creating this opportunity: UK net-zero farming, EU Sustainable Use Directive, UK path to sustainable farming.
The NemaRecognition project will showcase the feasibility and applicability of this technology toward PPN detection and would also represent proof-of-concept for developing similar innovations for other soil-dwelling organisms, with significant potential in the growing area of soil health services.
Technical Summary
Plant parasitic nematodes (PPNs) are destructive pests of many crops in temperate, sub-tropical, and tropical agricultural systems. Damage caused by their feeding results in a 14% loss in global crop yield. Furthermore, PPNs, can form damaging disease complexes with fungal and bacterial pathogens and act as virus vectors. Decisions on management, such as nematicide application, biopesticide use, choice of resistant/tolerant varieties and rotation length is frequently based on soil testing to determine the density of genera or species present. Following soil sampling, the extracted nematodes are assessed using taxonomic characteristics and dichotomous keys. Molecular tests such as qPCR may also be used in nematode diagnosis but these are mainly used to identity individual species of interest. To determine the array of free living PPNs in a soil sample, trained taxonomists, typically with at least 3-4 years of experience, are needed. Unfortunately, such specialists are in decline and there are few training providers. Cyst nematodes, consisting of the economically important genera Globodera and Heterodera spp., are routinely assessed in crops such as potatoes and sugar beet. The potato cyst nematodes (Globodera rostochiensis and G. pallida) occur in 48% of the land used for potato production in England and Wales and are associated with annual losses amounting to £26-50M. Current assessment of potato cyst nematode is mainly based upon the morphological characterisation of cyst shape (female nematodes) by competent analysts. This project will provide a transformative solution for this technical skills gap by developing, training, testing, and benchmarking a machine learning based automatic image recognition technique that will act as an alternative to the standard PPN identification system. This is the first time that machine learning has been adopted for the identification of multiple economically important PPN genera from soil. The first step in the process involves se
People |
ORCID iD |
| Po Yang (Principal Investigator) | |
| Matthew Back (Co-Investigator) |
Publications
| Description | This project represents a significant breakthrough in the field of agriculture, specifically in combating nematodes - tiny pests that can cause substantial damage to crops. Nematode pests affect agriculture worldwide, leading to billions in economic losses annually. The dataset's contribution to developing robust detection methods has a global impact, offering a scalable solution that can be adapted and used in diverse agricultural settings across the globe. Here's a simplified breakdown of what has been achieved through this project: 1. Creation of a Specialized Nematode Detection Dataset: We gathered a vast collection of images (over 1.3k) showing four different types of nematodes that significantly affect agriculture. This collection, with expert annotations for nematode classification, serves as a foundation for training and testing the AI. It ensures that the AI can accurately recognize and differentiate between various nematode species. The nematode detection dataset has significantly contributed to a deeper understanding of nematode diversity, providing valuable insights into the behavior and characteristics of different nematode species. 2. Development of an Automatic Nematode Detection Model: Utilizing the collected images, an AI (Artificial Intelligence) model was developed to identify nematodes in these images. Impressively, the model achieved a high accuracy of 96.4%, demonstrating the AI model's ability to automatically and accurately classify and count the nematodes present in the images. This high level of accuracy enables farmers and agronomists to identify nematode infestations early and accurately, leading to more effective and targeted interventions that protect crop health and yield. For students and professionals in agronomy, pathology, and related fields, this AI model and dataset serve as educational tools, providing hands-on experience in artificial intelligence and pest detection. They offer real-world applications of machine learning and encourage the development of skills in this cutting-edge technology area. 3. Introduction of Explanation Methods: To make the AI's decision-making process transparent and trustworthy, two innovative methods were introduced. The first method uses heatmaps to visually highlight areas in the images that the AI focused on for detection. The second method provides examples of similar images that the AI has learned from, offering insights into why it made specific decisions. These approaches help users better understand and trust the AI's detections. 4. Launch of a User-Friendly Website: Finally, all these advancements have been made accessible through a dedicated website. This online platform allows users, such as farmers and researchers, to easily upload images and receive instant nematode detection results. The website not only provides a practical tool for detecting nematodes but also offers interface services for integrating the nematode detection ability into other agriculture tools. In essence, NemaDetect AI has revolutionized how we detect and understand nematodes in agricultural settings, offering a highly accurate, transparent, and user-friendly tool for managing these pests more effectively. |
| Exploitation Route | The outcomes of the NemaDetect AI model and dataset can be advanced and utilized across both academic and non-academic sectors in several impactful ways. Academically, researchers in computational biology, agronomy, and artificial intelligence can utilize the dataset to refine AI models for pest detection, explore nematode behavior and diversity, or develop new methodologies for data analysis and interpretation. This research could lead to publications, collaborative projects, and educational resources, enhancing understanding and fostering innovation in pest management technologies. In the non-academic realm, agricultural technology companies and developers can integrate the NemaDetect AI model into their products and services, offering advanced nematode detection tools to farmers and agronomists worldwide. This application would directly contribute to more sustainable and effective pest management practices, ultimately increasing crop yields and reducing economic losses due to pest infestations. Additionally, governmental and non-governmental organizations focused on agriculture and food security might adopt and promote these AI-driven tools as part of their initiatives to support farmers in resource-limited settings, thereby enhancing global food production and sustainability. |
| Sectors | Agriculture Food and Drink |
| URL | http://13.49.74.210/ |
| Description | Various wider impacts would also be achieved: Economic: We estimate that this tool would process samples twice as fast as a nematologist. A faster service will result in cost savings for the service provider and lower end-user costs; potentially translating to a 30-40% saving. Many crops suffer from PPNs but are not screened due to perceived low risk and high testing costs. A cheaper testing service would encourage growers to test per ha and in more crops. This would improve PPN management, crop yields, and business productivity. This project presents a solution to the technical skills shortage facing plant testing laboratories. Without addressing the skills shortage, the industry has lower capacity to monitor and control PPN. This tool would reduce production costs for growers: cheaper services increase the likelihood that more assessments will be carried out. Nematacide costs are £450/ha (£3,600 per average field). If sampling per ha showed only 60% of the field required nematicide application this would represent a cost saving (including deduction of increased testing costs) of £880. Environmental: Improving PPN diagnosis will help growers guide chemical application and encourage non-chemical control. This may result in fewer or more nematicides being applied, but use would be better justified. Such approaches are in-line with the EU's Sustainable Use Directive, which has been transposed into UK law. Social: This tool could train nematologists at research and commercial testing laboratories. It is envisaged that this tool would not replace nematologists but increase the availability and quantity of testing. This would enable nematologists to concentrate on novel research, including sustainable control methods. Reduced production costs would lower food costs, thereby increasing affordability and food security. |
| First Year Of Impact | 2024 |
| Sector | Agriculture, Food and Drink |
| Impact Types | Cultural Societal Economic |
| Title | Microscopic images of nematodes for nematode detection |
| Description | The Nematode Detection Dataset is a comprehensive collection of 1,368 high-quality microscope images specifically curated for the advancement of agricultural pest management through machine learning. This dataset has been meticulously assembled to aid in the detection, identification, and analysis of four key types of nematodes that are critical to global agriculture: Meloidogyne (Root-knot nematodes), Globodera pallida (Potato cyst nematodes), Pratylenchus (Root-lesion nematodes), and Ditylenchus (Stem nematodes). Furthermore, it encompasses two significant life stages of nematodes: the Cyst stage and the Juvenile 2 (J2) stage, providing a diverse range of data for nematode detection model training and testing. Furthermore, corresponding nematode detection annotations are attached for data analysis. The primary goal of constructing this dataset is to: Enhance Detection Accuracy: Provide a rich source of labeled data to train deep learning models, improving their accuracy in detecting and classifying nematode pests in agricultural settings. Support Agricultural Research: Aid researchers and agronomists in studying nematode infestations, their impact on crops, and developing effective management strategies. Promote Precision Agriculture: Facilitate the development of AI-driven tools for precision agriculture, enabling farmers to take targeted actions against nematode threats, thereby optimizing crop health and yield. Currently, this dataset is not published. It will published in 2025 for use by the academic and research community, AI developers, and agricultural technology companies. Access policies and licensing details are structured to promote wide-ranging utilization and collaboration, aiming to accelerate innovation in agricultural pest management through advanced AI methodologies. DOI: 10.21227/rjbq-4a58 . |
| Type Of Material | Database/Collection of data |
| Year Produced | 2023 |
| Provided To Others? | No |
| Impact | The development of the Nematode Detection Dataset has had a profound impact on several fronts, notably in agricultural technology, academic research, and the advancement of precision agriculture. Here are some of the key impacts: Advancements in AI-based Pest Detection: The dataset has significantly contributed to the development and refinement of machine learning models specialized in the detection and classification of nematodes. By providing a rich, annotated dataset, researchers and developers have been able to train more accurate and efficient algorithms, leading to the creation of robust AI tools for real-time nematode detection. This marks a substantial improvement over traditional manual identification methods, offering faster, more reliable results. By this dataset, we constructed a detection model baseline for nematode detection, which achieved 97.6% mean average precision for nematode detection. Promotion of Precision Agriculture: The practical application of AI models trained on this dataset has bolstered the practice of precision agriculture. Farmers and agronomists now have access to tools that can detect nematode presence and assess infestation levels with high precision, allowing for targeted interventions. This capability helps in minimizing unnecessary pesticide use, reducing environmental impact, and optimizing crop health and yields. Cross-disciplinary Collaborations: The dataset's creation and application have fostered collaborations between computer scientists, agronomists, and biologists, promoting a multidisciplinary approach to solving agricultural challenges. These collaborations have spurred innovation, leading to the development of integrated pest management solutions that are more effective and environmentally friendly. Educational Impact: Finally, the dataset serves as an excellent educational resource for students and professionals interested in machine learning, computational biology, and agriculture. It provides a practical dataset for training and testing, enhancing learning opportunities and fostering a new generation of researchers and practitioners equipped with the skills to tackle agricultural pests through technology. |
| Title | NemaDetect AI (Website) |
| Description | The NemaDetect AI website serves as the central hub for accessing the innovative nematode detection and analysis service powered by advanced artificial intelligence. Designed with user experience in mind, the website offers a clean, intuitive interface that allows users-from farmers and agronomists to researchers and educators-to easily navigate and utilize the various features of the service. |
| Type Of Technology | Webtool/Application |
| Year Produced | 2023 |
| Impact | The development of NemaDetect AI has marked a significant milestone in the field of agricultural technology, bringing forth notable impacts on several fronts: Advancements in Precision Agriculture NemaDetect AI has propelled the capabilities of precision agriculture by providing an AI-powered tool for the accurate detection and classification of nematodes. This advancement enables farmers and agronomists to make informed decisions, significantly improving pest management strategies. By facilitating early detection and identification of nematode species, NemaDetect AI contributes to the sustainable management of crop health, reducing the reliance on chemical pesticides and minimizing environmental impact. Adoption and Usage Since its introduction, NemaDetect AI has seen widespread adoption among various stakeholders in the agricultural sector, including small-scale farmers, commercial agriculture operations, research institutions, and agronomy consultants. Its ease of use, combined with the power of AI to provide quick and accurate analyses, has made it a valuable tool across diverse agricultural settings worldwide. The software has been instrumental in enhancing crop yields and quality by enabling targeted nematode management practices. Impact on Crop Health and Yield Directly, NemaDetect AI has contributed to the improvement of crop health and yields by enabling the early detection of nematode infestations, which are among the most challenging pests to manage in agriculture. By identifying infestation levels and nematode species, users can apply specific and effective treatment strategies, significantly reducing crop damage and loss. This targeted approach has led to increased crop yields and quality, contributing to food security and economic stability for farmers. Contribution to Research and Education Indirectly, NemaDetect AI has impacted the academic and research community by providing a tool that supports the study of nematode behavior, infestation patterns, and management strategies. It has become a resource for educational institutions, offering students and researchers a practical application of AI in agriculture. This has not only contributed to the advancement of agricultural research but also fostered innovation in pest management solutions. Environmental and Economic Benefits NemaDetect AI's contribution to reducing the overuse of pesticides aligns with environmental conservation efforts, protecting soil health and water resources while ensuring the sustainability of farming practices. Economically, the tool has provided cost-saving benefits to users by optimizing the use of resources and reducing the financial losses associated with crop damage, thereby enhancing the profitability of agricultural operations. In conclusion, NemaDetect AI's development and deployment have made significant strides in revolutionizing pest management within agriculture. Its broad adoption and the tangible benefits it delivers to users underscore its impact on improving agricultural productivity, environmental sustainability, and economic viability for those involved in farming and crop management. |