Artificial intelligence applications, in particular, deep learning techniques, that underpin image recognition are transforming society. They are now widely used in image analysis for a wide range of applications requiring object segmentation, classification and recognition. Teams at Lincoln have pioneered there use in medical imaging (especially retinal images to detect eye disease) and pest recognition. More recently the Lincoln group have developed a highly accurate system to recognise, count and geotag locust insects in China. This system has deployed state of the art ResNet convolutional neural networks for high-speed pest recognition and these have also been enabled for application on a standard smartphone. This technology enables a step change in pest and diseases recognition and quantification, providing novel and powerful tools to assist agronomic support for multiple fruit pest and diseases.
The development of these systems is not trivial. They require very amounts of training data and input from expert agronomists. Therefore, in this project, we are developing a deep learning image library and algorithms to identify, enumerate and geo-locate a range of critical pests and diseases that impact soft fruit production, including SWD, Western Flower Thrips, aphids, powdery mildew, and several fruit rot fungal pathogens. This enables accurate pest and disease monitoring, at least as accurate as human agronomists, automation of agronomic support and more effective decision-making by fruit growers. A simple application would include pest counting and recognition on standard sticky traps as well as more challenging applications to recognise live pests on crops. Specific attention will be paid to investigate whether it is possible to detect those strawberry leaves or fruits with latent fungal infection.