Detection Of Crop Disease Using MobileNet

Authors:

Anshul Bhatt¹, Akshat Rawat², Ayush Semwal³, Ekta Uniyal⁴, Mr. Pradeep Chauhan⁵

Student, B-Tech-CSE, Shivalik College of Engineering, Dehradun, India, Email id-

bhattanshul001@gmail.com¹

Student, B-Tech-CSE, Shivalik College of Engineering, Dehradun, India, Email id

akshat5rawat@gmail.com²

Student, B-Tech-CSE, Shivalik College of Engineering, Dehradun, India, Email id

ayushsemwal724@gmail.com³

Student, B-Tech-CSE, Shivalik College of Engineering, Dehradun, India, Email id

uniyale02@gmail.com⁴

Assistant Professor, Shivalik College of Engineering, Dehradun, India, pradeep.chauhan@sce.org.in⁵

Abstract— Crop diseases hit harvests hard, especially when farmers can't spot them early. We’ve been looking at a new deep learning approach that uses MobileNetV2 and mixes structured images with real field photos to catch problems faster. The data covers four classes: Early Blight, Late Blight, Leaf Mold, and healthy leaves.

Preprocessing steps like resizing, normalization, augmentation, and cleaning out duplicates really helped the model hold up better. Their MobileNetV2 version hit 91.53% validation accuracy and dropped the loss to 0.235. The training curves stayed steady, and overfitting stayed low. Compared to regular CNNs, this setup gives solid accuracy without needing heavy computing power, which matters when you want something that works right in the field.

Keywords—Crop Disease Detection, MobileNetV2, Deep Learning, Transfer Learning, PlantVillage, PlantDoc, Agriculture AI.