A novel algorithm in artificial intelligence to enhance the precision of detecting plant diseases.

Every year, plant diseases caused by bacteria, viruses, and fungi contribute to significant economic losses. Rapid detection of these diseases is essential to limit their spread and mitigate agricultural damage, but it is a major challenge, especially in areas of high production. Smart cultivation systems use camera monitoring equipped with artificial intelligence (AI) models to detect plant disease traits, which often manifest as changes in leaf morphology and appearance.

However, conventional methods of image classification and pattern recognition extraction feature a reference to diseased plants from the training set. As a result, they have low interpretability, which means that it is difficult to describe the learned features.

Moreover, having large data sets to train the model is tedious. Handcrafted features, which are determined based on feature detectors, descriptors, and vocabularies designed by experts, offer a practical solution to this problem. However, this often leads to the adoption of irrelevant features, which reduces the performance of the algorithm.

Fortunately, a solution is now on the horizon. A team of data scientists and plant phenomenology experts from China and Singapore have developed an intelligence swarm feature selection (SSAFS) algorithm that allows effective image-based detection of plant diseases. They reported the development and validation of this algorithm in their recent study published in Plant Phenomenology.

Explaining the benefits of introducing SSAFS, the corresponding author of this study, Professor Ziwei Ji, commented, “SSAFS not only significantly reduces the number of features, but also greatly improves classification accuracy.”

The study used a combination of two principles: high-throughput phenomenology, in which plant traits such as disease severity can be analyzed on a large scale, and computer vision, in which image features that are representative of a particular condition are extracted. Using SSAFS and a set of plant images, the researchers identified a “subset of optimal traits” for plant diseases.

This subset included a list of only high-priority characteristics that could successfully classify a plant as diseased or healthy, and further estimate disease severity. The effectiveness of SSAFS was tested in four UCI data sets and six vegetative phenotypes data sets. These datasets were also used to compare the performance of SSAFS to that of five other similar swarm intelligence algorithms.

The results show that SSAFS performs well in both detecting plant diseases and estimating their severity. In fact, it outperforms the latest algorithms out there in identifying the most valuable features of handcrafted photos. Interestingly, the majority of these disease-related features were local—that is, they included distinct patterns or structures, such as dots, edges, and spots, that are often observed in diseased plants.

Overall, this algorithm is a valuable tool for obtaining an optimal combination of handcrafted image features indicating plant pathologies. Its adoption can greatly improve the recognition accuracy of plant diseases and reduce the processing time required.

When asked about the future implications of their study, Professor Ji said: “One of the critical contributions of this work to phenomenological implantation is the definition of man-made features and the accurate screen of relevant traits through a new computational approach. We propose to combine comprehensive man-made and non-man-made features of plant images from For accurate and effective detection in the field of phenomena.”