By Andrew Porterfield
Corn is an economically important crop in Asia, Europe, and the Americas. But consumers of corn are not limited to humans and livestock. The crop is also subject to consumption by more than 30 species of insects in the order Lepidoptera (i.e., moths and butterflies). Four major pests are the most destructive: Ostrinia furnicalis, Spodoptera frugiperda, Mythimna separata, and Spodoptera litura.
The destructive capabilities of these insect pests vary according to their larval stages, or instars. Therefore, it is important for growers and pest management professionals to be able to identify stages of each pest to optimize their management. Traditionally, identification is done by the naked eye and knowledge of each pest. This can be slow and laborious and can mis-identify lepidopteran stages. There are very small size differences between stages, and larvae of different species share similar colors.
A research team at Jilin Agricultural University in China recently looked at whether certain artificial intelligence (AI) models could identify the different stages of these insects more quickly and accurately. After testing several AI models, they found one combination of AI model and data optimizer that was the most effective for classifying insects by age in 23 developmental stages of the four pests. The researchers found that this AI/machine learning combination had an overall accuracy of 96.65 percent, outperforming all the other models they tested. Their findings were published in October in Environmental Entomology.
By leveraging machine learning algorithms, farmers and researchers can make identification faster and easier and detect pests earlier, allowing for more precise control measures and enhancing both crop protection and sustainability. However, it’s important to use the right AI model.
One major issue in applying AI to insect identification is being able to manage images from different sources and of different sizes and transforming this raw data into a usable and shareable representation of insect features, such as shape, geometry, color, and texture.
Ri-Zhao Chen, Ph.D., and his team at Jilin Agricultural University used convolutional neural network (CNN) models, which can classify and detect images better, with less memory and computation needed. They also applied transfer learning, which relies on pre-trained models from previous machine-learning tasks and further reduces the need for advanced (and expensive) computation.
In their study, the researchers selected larvae from the four top destructive lepidopteran corn pests (O. furnicalis, S. frugiperda, M. separata, and S. litura). Larvae were served with corn leaves as their only food source. Twenty larvae were selected for rearing. The first through fourth instar larvae of all four species were photographed with a microscope, while the fifth instar of O. furnicalis and fifth and sixth instar of the other three species were photographed with a mobile camera. Every 24 hours, the researchers took 20 images of each pest. In the field tests, the researchers introduced insect larvae to an existing corn field, where images were taken using a mobile phone.
The key task for the AI, then, was to resolve all this disparate data and correctly identify each species and instar. Of five CNN models, one called “Densenet121” stood out as the most accurate. But the model depended on data augmentation software that could harmonize the photos and micrographs. An optimizing algorithm named “Adam” was shown to be the most accurate.
Artificial intelligence in agriculture is not a new phenomenon, but the researchers say the real challenge lies in finding the best model in a “plethora of AI-driven approaches.”
Click here to see more...