Directed acyclic graphs-based diagnosis approach using small data sets for sustainability

Abstract

Improving the ability of crop disease diagnosis in agriculture to reduce water pollution caused by pesticide abuse plays a key role in realizing the model of “sustainable intensification”. The current common methods of disease diagnosis through machine vision need an extremely large image data set to reduce the interference of light, background, and other factors. Unfortunately, even common crops will have the problem of lack of suitable and sufficient disease image data sets. In this paper, a disease model based on graph theory is designed to solve the problem of large data sets required by machine vision methods. The model adds the concept of guidepost to the standard directed acyclic graph, which solves the problem of mutual interference between the paths of different diseases. Based on the disease model, the disease diagnosis algorithm designed in this paper achieves effective disease diagnosis under the missing input indicator sequence. Based on the above, a disease diagnosis system with B/S structure is constructed. The system includes two core modules: knowledge base and inference engine. Finally, based on the disease diagnosis system, the disease diagnosis algorithm is tested. The test takes the disease diagnosis of watermelon as an example. Under the scenario of complete and partial missing input indicator sequence, the specific cases are divided into 11 kinds, and three tests are carried out respectively. The results show that the algorithm not only reduces the dependence on data sets, but also can still achieve effective disease diagnosis through fuzzy inference when the input indicator sequence is chaotic and missing. Due to its low data set requirements, the results of this study have good applicability, which will greatly improve the intelligent development of crop disease diagnosis in agriculture.