Application of a multi-layer convolutional neural network model to classify major insect pests in stored rice detected by an acoustic device

Studies reported that 12-40% of stored grains are lost due to insects, but the use of early detection devices such as acoustic sensors can guide subsequent storage management to reducing losses. Acoustic detection can directly identify the cause of damage (i.e., insects) in stored grains rather than the effect (e.g., RH, temperature) and it is capable of handling high information density due to the broad frequency band and the different sound levels. This research addresses the question if the use of micro-electromechanical system (MEMS) microphone can detect insect sound in stored grains, predict insects' presence and classify insects according to species with the application of a multi-layer convolutional neural network (CNN) algorithm. We adapted the acoustic sensor from the Smart Apiculture Management Services (SAMS) project using the Adafruit SPH0645, an inexpensive MEMS microphone that was used to detect insect pests in stored rice grain. The recorded sounds of major insect pests (adult stage) in stored paddy grains, namely, lesser grain borer (Rhyzopertha dominica, Fabricius), rice weevil (Sitophilus oryzae, Linnaeus), and red flour beetle (Tribolium castaneum, Herbst) were characterized using spectrogram profiles. Machine learning technique was applied using CNN with an average accuracy of 84.51% to classify insect pests from the emitted sound profiles. The use of an acoustic detection system and the application of a CNN classification model provides an efficient method of detecting hidden insects in stored grains that can guide farmers and end-users in implementing appropriate and timely insect pest control without applying harmful chemicals in stored grains.