Research Area: Machine Learning
Deep learning is a promising approach for fine- grained disease severity classification for smart agriculture, as it avoids the labor-intensive feature engineering and segmentation-based threshold. In this work, we first propose a Densely Connected Convolutional Networks (DenseNet) based transfer learning method to detect the plant diseases, which expects to run on edge servers with augmented computing resources. Then, we propose a lightweight Deep Neural Networks (DNN) approach that can run on Internet of Things (IoT) devices with constrained resources. To reduce the size and computation cost of the model, we further simplify the DNN model and reduce the size of input sizes. The proposed models are trained with different image sizes to find the appropriate size of the input images. Experiment results are provided to evaluate the performance of the proposed models based on real- world dataset, which demonstrate the proposed models can accurately detect plant disease using low computational resources.
Keywords:
Deep Learning
Plant Disease Detection
Smart Agriculture
Machine Learning
Densely Connected Convolutional Networks (DenseNet)
Author(s) Name: Laha Ale; Alaa Sheta; Longzhuang Li; Ye Wang; Ning Zhang
Journal name: IEEE Globecom Workshops (GC Wkshps)
Conferrence name:
Publisher name: IEEE
DOI: 10.1109/GCWkshps45667.2019.9024439
Volume Information:
