Original papersAutomated vision-based system for monitoring Asian citrus psyllid in orchards utilizing artificial intelligence
Introduction
The Huanglongbing (HLB), or citrus greening, is a significant disease that affects citrus orchards causing rapid decline of the trees (Chung and Brlansky, 2005). HLB was first reported in 1919 in southern China, and it is spread already to 40 different countries all around the world, including the USA (Bové, 2006). The disease causes an immunologic resistance reduction affecting orange production by causing fruit drop and the production of small, misshapen and low-level of juice fruit with no economic value (Hodges and Spreen, 2012, Albrecht and Bowman, 2009). The causal agent of HLB in the America and Asia is usually the Candidatus Liberibacter asiaticus (CLas), a phloem limited gram-negative bacterium (Jagoueixet al., 1994, Garnier and Bové, 1996). In 17 years (between 2000 and 2017), the citrus cultivated area in Florida declined 45% and the volume production utilized declined by 71%, mainly due to problems generated by the HLB (Court et al., 2018). Between 2012 and 2017 in Florida, citrus production revenue decreased approximately 47% (from approximately 2.1 billion dollars to approximately 1.2 billion dollars) (Court et al., 2018), resulting in an average annual reduction of 7945 jobs and $1.098 billion in industry output (Court et al., 2017).
The HLB is transmitted by the vector Diaphorina citri Kuwayama, most know as Asian Citrus Psyllid (ACP) (Hemiptera: Phyllidae). It was verified in Florida for the first time in 1998 (Halbert and Manjunath, 2004). The transmission of the disease occurs in three steps: (i) an initial period of acquisition, when the nymphs and the adults come in contact with the pathogenic agent (CLas), (ii) a latency period, that may include a bacteria reproduction too (Inoue et al., 2009) and (iii) the inoculation period, where the vector transmits the bacteria (presents in the salivary glandules, muscles, tissue and ovaries) to the plant (Ammaret al., 2011, Ammaret al., 2010). The ACP can move about 100 m in 3 days (Boina et al., 2009) and recent data showed a dispersion distance of at least 2 km in 12 days (Lewis-Rosenblum, 2011). This vector exhibits biological characteristics such as high reproduction capacity, fast population growth ratio and capacity to withstand a wide temperature range, which allows it to spread quickly (Halbert and Manjunath, 2004). Bacteria control is becoming virtually unsustainable as robust methods of bacteria elimination are expensive and not effective (Halbert and Manjunath, 2004). Therefore methods for monitoring and controlling the ACP populations are shown to be more advantageous in the prevention and mitigation of the citrus greening problem, allowing the treatment only in critical plants, saving time and reducing costs (Halbert and Manjunath, 2004).
A method to combat the ACP vector includes extensive chemical control programs, by the application of pesticides. However, empiric studies are showing that pesticides application to prevent the introduction and dissemination of ACP has not been very effective (Gottwald, 2007). Besides that, ACP populations have become more resistant to chemicals (Tiwari et al., 2011), which can become problematic without controlled use of pesticides. The use of biological agents as natural predators of the vector is another method of population control (Hall, 2013), in which case chemicals can reduce the effectiveness of the method (Halbert and Manjunath, 2004). Fungi control (entomopathogenic fungi) was also reported as a good tool to control ACP populations (Hallet al., 2013, Moranet al., 2011, Samson, 1974, Subandiyahet al., 2000). All the mentioned control methods should be supported by a strong geolocation analysis of ACP quantities, identifying most affected areas and generating accurate targets, both to chemical and biological control of the vector.
Monitoring the ACP population is an essential component of ACP management, both for application of economic thresholds as well as assessing effectiveness of control actions (Monzo and Stansly, 2017). For this purpose, the tap sample method (other traditional ACP monitoring methods are presented below), which requires striking a randomly selected branch and counting ACP falling onto a sheet, has proven to be an efficient and reliable tool for assessing ACP numbers in the tree canopy (Hall and Hentz, 2010, Monzoet al., 2015). Spraying based on need as indicated by tap sample counts has been shown to reduce ACP management costs and conserve natural enemies (Monzo et al., 2014, Monzo and Stansly, 2017). However, this tap sample manual counting method is very labor intensive and time consuming.
Since labor shortage is a major issue in USA, machine vision techniques, internet of things (IoT) (Ampatzidiset al., 2018, Ampatzidiset al., 2012) and cloud-based technologies (Ampatzidis et al., 2016) can simplify pest scouting procedures and improve precision spraying applications (Partel et al., 2019), reduce labor cost, decrease data collection time, and produce critical and practical information (Ampatzidis et al., 2017, Luvisi et al., 2016). Rapid methods for early detection of pests and diseases would assist growers in making timely management decisions and to limit spread (Abdulridhaet al., 2019, Abdulridha et al., 2018).
Machine vision along with artificial intelligence (AI) techniques has been increasingly applied to agriculture (Cruz et al., 2017). Moller (2010) concluded that using computer vision technologies on agricultural operations lowers the operator stress levels (Moller, 2010). Deep learning based convolutional neural networks (CNN) is the most common AI approach for image recognition and have proven to achieve great performance on image detection and classification tasks (Krizhevskyet al., 2012, Güçlü and Van, 2014, Cruzet al., 2019). Their deep architecture and good weight equalization schemes provide great sensitivity to detect complex and high level features. CNN’s have the advantage to be trained from large sets of data, eliminating the need to manually design feature extraction algorithms. In this study, a novel artificial intelligence technology utilizing deep learning was developed and evaluated to automate ACP scouting procedures in citrus orchards, providing rapid and valuable information to improve ACP management.
Section snippets
Traditional ACP monitoring methods
There are several methods to monitor ACP populations in order to determine the need to spray (Stanslyet al., 2010, Hallet al., 2007): (i) the yellow sticky trap which uses an adhesive board to collect psyllids for later visualization; however, sticky traps sampling is slow, labor intensive, costly and assesses ACP in flight, which may not always correlate well with numbers in trees. (ii) Sweep nets, where a net of 15-inch diameter is used into the canopy of trees to collect insects; (iii) the
Experimental design
All the tests were performed at the University of Florida’s Southwest Florida Research and Education Center (SWFREC) located in Immokalee, Florida, USA. The grove used to evaluate the developed technology contains 8 rows of 80 m length with two different citrus varieties, Valencia and Swingle citrumelo, being 18 months old at the date of the experiment. The prototype was attached to an ATV (John Deere Gator™) to move it around the field. All the data were collected on February 13th of 2019,
Results and discussion
The results for the experiment on the 90 trees are shown in Table 1. A total of 267 true positive detections were counted, 69 false positives and 14 false negatives, which result in a precision and recall of 80% and 95%, respectively. The overall f-score result was 87%. The results show that the software performed better on ACPs sensitiveness (only 14 missed psyllids) than on accuracy (69 debris misidentified as ACPs), which shows that it was easier to detect ACPs than to distinguish an ACP
Conclusion and future work
A cost-effective automated system to detect, distinguish, count and geo-locate Asian citrus psyllid (ACP) in a citrus grove utilizing machine vision and artificial intelligence was developed and evaluated. This novel technology was designed to automate the conventional stem tap method for ACP scouting. It comprises a tapping mechanism to hit the tree’s branches so that insects fall over a board with a grid of cameras used for image acquisition. An NVIDIA TX2 embedded computational unit was used
References (44)
- et al.
A remote sensing technique for detecting laurel wilt disease in avocado in presence of other biotic and abiotic stresses
Comput. Electron. Agric.
(2019) - et al.
Development and evaluation of a novel system for monitoring harvest labor efficiency
Comput. Electron. Agric.
(2012) - et al.
Cloud-based harvest management information system for hand-harvested specialty crops
Comput. Electron. Agric.
(2016) - et al.
Detection of Grapevine Yellows Symptoms in Vitis vinifera L. with Artificial Intelligence
Comput. Electron. Agric.
(2019) - et al.
Development and evaluation of a low-cost and smart technology for precision weed management utilizing artificial intelligence
Comput. Electron. Agric.
(2019) - et al.
Isolation and characterization of two entomopathogenic fungi attacking Diaphorina citri (Homoptera, Psylloidea) in Indonesia
Mycoscience
(2000) - et al.
Evaluating the performance of spectral features and multivariate analysis tools to detect laurel wilt disease and nutritional deficiency in avocado
Comput. Electron. Agric.
(2018) - et al.
Candidatus Liberibacter asiaticus and huanglongbing effects on citrus seeds and seedlings
Horticult. Sci.
(2009) - et al.
Detection and relative titer of Candidatus Liberibacter asiaticus in the salivary glands and alimentary canal of Diaphorina citri (Hemiptera: Psyllidae) vector of citrus huanglongbing disease
Ann. Entomol. Soc. Am.
(2011) - et al.
Localization of Candidatus Liberibacter asiaticus, associated with citrus huanglongbing disease, in its psyllid vector using fluorescence in situ hybridization
J. Phytopathol.
(2010)
iPathology: robotic applications and management of plants and plant diseases
Sustainability
Voice-Controlled and Wireless Solid Set Canopy Delivery (VCW-SSCD) System for Mist-Cooling
Sustain., Spec. Issue: Inform. Commun. Technol. (ICT) Sustain.
UAV-based high throughput phenotyping in citrus utilizing multispectral imaging and artificial intelligence
Remote Sens.
Tap Sampling for Asian Citrus Psyllid (ACP) Field Sheet
Quantifying dispersal of Diaphorina citri (Hemiptera: Psyllidae) by immunomarking and potential impact of unmanaged groves on commercial citrus management
Environ. Entomol.
Huanglonbing: A destructive, newly-emerging, century-old disease of citrus
Plant Pathol.
Economic Contributions of the Florida Citrus Industry in 2015–16
Economic Contributions of the Florida Citrus Industry in 2016–17
X-FIDO: an effective application for detecting olive quick decline syndrome with novel deep learning methods
Front., Plant Sci.
Distributions of the Huanglonbing (Greening) Liberobacter Species in Fifteen African and Asian Countries
Citrus canker and citrus huanglongbing, two exotic bacterial diseases threatening the citrus industries of the Western Hemisphere
Outlooks on Pest Manage.
Cited by (76)
A smartphone application for site-specific pest management based on deep learning and spatial interpolation
2024, Computers and Electronics in AgricultureA deep learning-based pipeline for whitefly pest abundance estimation on chromotropic sticky traps
2023, Ecological InformaticsUnmanned aerial vehicle (UAV) imaging and machine learning applications for plant phenotyping
2023, Computers and Electronics in AgricultureDetecting two-spotted spider mites and predatory mites in strawberry using deep learning
2023, Smart Agricultural TechnologyUnderstanding the potential applications of Artificial Intelligence in Agriculture Sector
2023, Advanced AgrochemDiagnosis of grape leaf diseases using automatic K-means clustering and machine learning
2023, Smart Agricultural Technology