A review on the use of artificial intelligence and deep learning algorithms in crops Phytosanitary Monitoring

Abstract

The article presents a review of current applications and capacities of artificial intelligence in identifying pests and diseases of common agricultural crops. The review is created based on the literature search, conducted using Google Scholar engine. In total, 42 recent scientific papers (22 published in journals and 20 conference proceedings), which were published within last five years, were analysed and included into this review. Scientific papers providing incomplete data, controversial and biased information were excluded from the review. Mainly open access sources were included into the review. As a result, it was established that current artificial intelligence applications in phytosanitary monitoring of crops allow a great decrease in the mistakes in pests and diseases identification, as well as a reduction in the expenses of labour and time for manual observations. Generally, most AI-based models for plants pests and diseases identification provided identification accuracy and specificity exceeding 85%-90%, some of them reaching the peak of 99%-100% accuracy. The best results are mainly recorded for convolutional neural networks and their combination with other machine learning techniques. However, no clear unified algorithm is recommended for this purpose. The best results are usually associated with neural network-based algorithms. Further deeper scientific research is necessary to clarify the advantages, drawbacks and pitfalls of artificial intelligence application in phytopathological surveys.