AI-DRIVEN PEST DETECTION SYSTEMS: INTEGRATING ZOOLOGICAL KNOWLEDGE AND AGRICULTURAL ENGINEERING

Abstract

Artificial intelligence has created new opportunities to deal with the pests in the farming area accurately, and the majority of the detection system does not consider the environment.  This research is a demonstration of an integrated pest detection software supported by AI that incorporates knowledge in zoology and agricultural engineering in order to make it more precise and less to understand.  The UAVs captured high-resolution images of crops infected with pests, and the entomologists assigned labels to them creating some powerful dataset with more than 50 species.  The idea was to test and train a ResNet50 convolutional neural network (CNN) 5-fold cross-validation. It consisted of a semantic layer of attention which linked morphological characteristics to entomological expertise.  The model presented an accuracy of classification of 94.3 percent and an AUC of 0.97. The outputs were quite comprehensible and revealed characteristics which were exclusive to pests.  Real-time field deployment of ground vehicles enabled by edge computing demonstrated that the system performed well in the real world scenario, in terms of its capabilities to identify things.  Comments left by zoologists helped us to enhance the attention mechanisms and fill the model with whatsoever blind spots.  The technology also worked in most environmental conditions, though very good at separating similar-looking pests.  The multidisciplinary approach indicates how a combination of a deep learning algorithm with the specificity of biological intelligence will result in a pest detection solution capable of next-generation smart farming which can be deployed in field, scale, and explainable.