Automatic detection of insect predation through the segmentation of damaged leaves

Abstract

Leveraged by the production of grains, oilseeds, and fresh deciduous fruits, food production has reached new heights, exceeding the amount produced in previous years and with an estimate of new records for the coming years. In this sense, technological advances are essential to reduce costs and increase quality and productivity. In this paper, we present a novel method to detect insect predation on plant leaves that uses geometric leaf properties and digital image processing techniques to construct image models. Unlike other approaches, our method detects and highlights the regions of leaves attacked by insects and segments the contours of insect bites. We evaluated our proposal considering 12 crucial crops for the world market, and it demonstrated to be effective, even in the presence of noise, image scale, and rotation. Besides, it identifies insect predation areas regardless of the plant species with precision above 90% in blueberry, corn, potato, and soybean leaves. Thus, this proposal introduces a new approach to automatic leaf analysis and contributes to reducing human effort in identifying the occurrence of pests. The code prepared by the authors is publicly available.

Keywords

• Leaf segmentation
• Foliar analysis
• Insect predating
• Smart farming
• Precision agriculture