The success achieved with the development of image recognition software for identifying individuals by facial features, and diagnosing melanomas and other skin disorders, has prompted several applications of machine deep-learning to the identification and diagnosis of “pests”.
The development of an AI for the identification of the brown marmorated stink bug in New Zealand, mentioned earlier, is one example of the potential of AI to address a specific biosecurity issue. Intended for use by citizen scientists, the aim is to achieve rapid identification of this important invasive pest at high risk of entry to New Zealand. Another prototype app, for the identification of weed seeds, uses the camera on a smartphone [Schmidt-Lebuhn et al. 2020]. This app provides the user with estimates of confidence in an identification, species profiles, and thumbnail images.
Other examples of AI applications to plant pests are concerned with the diagnosis of general insect pest problems or diagnosing specific crop diseases. Plantix is a mobile crop advisory app for farmers, extension workers and gardeners. It provides diagnostic services on pest damage, plant diseases, and nutrient deficiencies affecting a wide range of crops and suggests suitable treatment measures. Plantix currently combines the results of AI diagnostic technology with support from an online community of scientists, farmers, and plant experts. Plantix will modify its modus operandi when the accuracy of the AI improves with more images being analysed.
Another crop diagnostic app that uses AI technology, and which also has a back-up expert community to respond to diagnostic requests, has been developed by the PlantVillage community plant protection support team at Penn State University. “Nuru” was created as an AI assistant for small holder farmers in Africa, with three AI components:
1. human expert-level crop disease diagnostics based on computer vision; 2. use of ground and satellite derived data to improve diagnosis; and 3. use of language comprehension and automated responses to farmers’ questions.
In China, there are several examples of AI platforms being used to provide plant protection support.
Love Plant Protection is an artificial intelligence platform for plant protection developed at Zhejiang University. It supports online intelligent identification of agroforestry diseases, insect pests, and weeds; remote control guidance; data analysis; and an agricultural information service. App users can take pictures on site and accurately identify more than 600 kinds of agroforestry diseases, pests, and weeds, and more than 200 plants online. Users can access reference information and obtain remote technical guidance from experts. The information provided by users can yield information about the real-time occurrence and spread of agricultural and forestry diseases, pests, and weeds in specific areas.
The social communication platform, WeChat, has been used by a scientific company in Beijing, to develop a crop diagnostic app. Users can photograph disease symptoms on seven specific crops (currently) and upload these images to the AI cloud. In return, they receive information about the disease causing the symptoms, the confidence level of this diagnosis, and fact sheets about the disease, including management options.
A project for diagnosing pests and diseases of banana using AI technology has been reported by the International Centre for Tropical Agriculture (CIAT). In this case, the plan is to use deep convolutional neural networks (DCNN) and transfer learning to develop an AI-based banana disease and pest detection system to support banana farmers.
Typically, the development of image-based, identification tools employing AI requires large numbers of images, often of sub-optimal or non-standard views. All tools of this type are challenged by the existence of large numbers of poorly characterised ‘background’ taxa and by cryptic or visually very similar species.
