Image-based orchard insect automated identification and classification method

  • Authors:

  • Chenglu Wen;Daniel Guyer

  • Affiliations:

  • Cognitive Science Department, Xiamen University, Xiamen 361005, China and Fujian Key Laboratory of the Brain-like Intelligent Systems, Xiamen University, Xiamen 361005, China;Department of Biosystems and Agricultural Engineering, Michigan State University, East Lansing, MI 48824, USA

  • Venue:
  • Computers and Electronics in Agriculture
  • Year:
  • 2012

Quantified Score

Hi-index 0.00

Visualization

Abstract

Insect identification and classification is time-consuming work requiring expert knowledge for integrated pest management in orchards. An image-based automated insect identification and classification method is described in the paper. The complete method includes three models. An invariant local feature model was built for insect identification and classification using affine invariant local features; a global feature model was built for insect identification and classification using 54 global features; and a hierarchical combination model was proposed based on local feature and global feature models to combine advantages of the two models and increase performance. The three models were applied and tested for insect classification on eight insect species from pest colonies and orchards. The hierarchical combination model yielded better performance over global and local models. Moreover, to study the pose change of insects on traps and the hypothesis that an optimal time to acquire and image after landing exists, advanced analysis on time-dependent pose change of insects on traps is included in this study. The experimental results on field insect image classification with field-based images for training achieved the classification rate of 86.6% when testing with the combination model. This demonstrates the image-based insect identification and classification method could be a potential way for automated insect classification in integrated pest management.