|
چکیده
|
The excessive use of chemical substances in potato cultivation, while raising production costs, may lead to the development of hereditary resistance in weeds to pesticides, environmental degradation, and risks to the health of farm workers and end consumers. Therefore, accurately and promptly identifying the position and growth stage of the potato plant, as well as the location and type of weeds in the field, is a crucial step toward achieving intelligent variable-rate precision fertilizer and pesticide application machines and robots for managing chemical use on farms. Consequently, in this research, a precise, rapid, and hardware-efficient metaheuristic algorithm was developed, employing a combination of two sequential models, YOLOv9t, and image processing techniques. Initially, RGB images were sent to the first model to identify the position and growth stages of the potato plant and the location of broadleaf weeds. Then, using image processing techniques, the positions of these plants were removed from the original image, and the resulting image was converted to the HSV color space. The H color channel, after thresholding, was sent as input to the second model to identify the positions of narrowleaf weeds. The evaluation results indicated that this 9.17 MB algorithm attained a detection speed of 59 fps and an accuracy of 96.4%, which was 9.2% higher than the baseline YOLOv9t model's accuracy. These findings demonstrate the significant potential of the proposed algorithm for use in intelligent variable rate precision potato spraying machines and robots.
|