Precision Agriculture Through Deep Learning: Tomato Plant Multiple Diseases Recognition with CNN and Improved YOLOv7

• Published in: IEEE access 2024-01, Vol.12, p.1-1
• Main Authors: Umar, Muhammad, Altaf, Saud, Ahmad, Shafiq, Mahmoud, Haitham, Mohamed, Adamali Shah Noor, Ayub, Rashid
• Format: Article
• Language: English
• Subjects: Agriculture; Blight; Computer viruses; Convolutional Neural Network; Convolutional neural networks; Crop yield; Crops; Data acquisition; Data points; Deep learning; Error analysis; Feature extraction; Image classification; Leaf mold; Leaves; Machine learning; Plant diseases; Plants (biology); SIFT; Signs and symptoms; Target detection; Tomato Leaf Diseases Detection; Tomatoes; Viruses; YOLOv7
• ISSN: 2169-3536; 2169-3536
• DOI: 10.1109/ACCESS.2024.3383154

The ability to accurately identify tomato leaves in a field setting is crucial for achieving early yield estimation, particularly with the growing importance of Precision Agriculture. It may be difficult to determine exactly what diseases are affecting tomato plants due to the overlap in symptoms between different diseases. These are the earliest signs of disease that we found in the leaves of tomato plants. Yellow leaf curl virus, leaf mold, light blight, early blight, Mosaic virus, Septoria leaf spot, and bacterial spot are just some of the seven types of plant leaf diseases that were taken into account in this paper. For the development of a testbed environment for data acquisition, the greenhouse at the university was utilized for data on the leaves of tomato plants. This study proposes a target detection model based on the improved YOLOv7 to accurately detect and categorize tomato leaves in the field. To improve the model's feature extraction capabilities, we first incorporate the detection mechanisms SimAM and DAiAM into the framework of the baseline YOLOv7 network. To reduce the amount of information lost during the down sampling process, the max-pooling convolution (MPConv) structure is then improved. After that, this model arrived at a satisfactory outcome. Then, the image is segmented using the SIFT technique for classification, and the key regions are extracted for use in calculating feature values. After that, these data points are sent to a CNN classifier, which has a 98.8% accuracy rate and a 1.2% error rate. Finally, we compare our study to previous research to show how useful the proposed work is and to provide backing for the concept.