Crop Monitoring via Deep-Learning-Based Pest Classification for Pesticide Reduction

The development of new plant diseases and the spread of dangerous pests have caused severe damage to the agricultural economy. Due to a non-specialist knowledge of the subject, the measure adopted by farmers has been an excessive and massive use of pesticides which resulted in severe environmental pollution. Therefore, the promotion of safe and effective monitoring methods is desirable and even necessary. Pest management is a key issue in the field of agriculture to boost yields and consequently local economy. From a food safety perspective, pest detection and control is certainly significant for producing healthy crops characterized by high-quality, pesticide-free fruits and vegs. Multi-class pest detection is still a challenging topic because of the similarity among different species. A further effort comes from the possibility that the same insect can manifest itself on crop at different stages of its life cycle such as egg, larva, pupa and adult. This paper proposes a real-time object recognition approach for large-scale multi-class pest detection and classification based on deep-learning. The proposal is validated by adopting the IP102 dataset and the obtained results show the method validity and the improvements compared to some relevant studies indicated in the literature.