A New Deep Vision-Based Identifier As An Intelligent Herbicide Spraying Agent For Potato Farm Application
Weeding in potato fields with the least consumption of inputs (herbicide or mechanical operations) is one of the main goals of sustainable agriculture. In precision agriculture, machine vision-based systems are used as on-the-go sensing units to detect weeds from the main plant. This paper presents...
Saved in:
| Main Authors: | , , , |
|---|---|
| Format: | Article |
| Language: | fas |
| Published: |
Semnan University
2024-12-01
|
| Series: | مجله مدل سازی در مهندسی |
| Subjects: | |
| Online Access: | https://modelling.semnan.ac.ir/article_9175_ee4ab5f4bcc7e0fac1b03951af1d4ce4.pdf |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| Summary: | Weeding in potato fields with the least consumption of inputs (herbicide or mechanical operations) is one of the main goals of sustainable agriculture. In precision agriculture, machine vision-based systems are used as on-the-go sensing units to detect weeds from the main plant. This paper presents a new approach based on deep learning to detect weeds in potato fields. For this purpose, first, a comprehensive database was created, including the acquired images of the potato field (at different stages of plant growth, at different distances of the camera from the ground, at different hours of the day, and in different environmental conditions). Then, the location of the plants in the field (including weeds and potato plants) was determined using the deep YOLOV3 algorithm. Finally, to separate weeds from the main plant as well as to determine the type of weed, three different types of convolutional neural networks were developed. The results showed that the YOLOV3 algorithm is well able to localize the plants in the images. EN-Inception-V3 classifier was able to distinguish weeds from potato plants in the set of test images with 99.42% accuracy. The classification results of 9 different weed species using the developed deep learning models were satisfactory; so that the overall accuracy of EN-Inception-V3, EN-VGG-16, and HCNN models was 99.82%, 99.89%, and 92.83% in the training phase, and 96.69%, 90.32%, and 82.67% in the test phase, respectively. It should be noted that the combination of models leads to 98.2% accuracy in detecting the type of weeds. |
|---|---|
| ISSN: | 2008-4854 2783-2538 |