Development of a weed detection system using machine learning and neural network algorithms

The detection of weeds at the stages cultivation is very important for detecting and preventing plant diseases eliminating significant crop losses, traditional methods performing this process require large costs human resources, in addition to exposing workers risk contamination with harmful chemicals. To solve above tasks, also order save herbicides pesticides, obtain environmentally friendly products, a program agricultural pests using classical K-Nearest Neighbors, Random Forest Decision Tree algorithms, as well YOLOv5 neural network, proposed. After analyzing geographical areas country, from images collected weeds, proprietary database more than 1000 each class was formed. A brief review researchers' scientific papers describing they developed identifying, classifying discriminating based on machine learning convolutional networks deep algorithms given. As result research, weed system architecture quality estimates were obtained. According results assessment, accuracy by classifiers 83.3 %, 87.5 80 %. Due fact that species differ resolution level illumination, network have corresponding indicators intervals 0.82–0.92 class. Quantitative obtained real data demonstrate proposed approach can provide good low-resolution weeds.