SEARCH RESULT

Agriculture is today a major player in the global economy and has been the foundation of societies throughout history. As agricultural systems become increasingly complex, crop diseases have become a major challenge, threatening productivity and food security. Recent technological advances have opened up new possibilities for addressing these issues, particularly with the use of convolutional neural networks (CNNs). Still, running CNN models on edge devices—like smartphones or compact processors—hasn’t been explored enough, even though such tools could be a game-changer for farmers in remote areas. They enable prompt, real-time diagnosis without the need for internet connectivity or extensive infrastructures. This study takes a closer look at four lightweight CNN models: MobileNetV2, SqueezeNet, ShuffleNetV2, and MnasNet. Using a dataset of diseased plant images, we tested each model’s accuracy, processing time, and resource demands. The aim is to find which model strikes the best balance between precision and performance on limited hardware. ShuffleNetV2 emerged as the most effective, achieving a validation accuracy of 99.35% and offering the highest efficiency for deployment on edge devices.

International Conference on Advanced Technologies, Computer Engineering and Science
ICATCES

Thabet Righi Mohammed Charaf Eddine Meftah