A method of detecting apple leaf diseases based on improved convolutional neural network.

Area of Science:

• Agricultural Science
• Computer Science
• Artificial Intelligence

Background:

• Apple tree diseases pose significant challenges to orchard farmers.
• Existing deep learning models for crop disease detection face limitations due to complex leaf structures and similar disease appearances.

Purpose of the Study:

• To develop a novel deep learning model, DF-Tiny-YOLO, for rapid and accurate automatic detection of apple leaf diseases.
• To improve upon existing detection models by enhancing accuracy and speed.

Main Methods:

• Proposed DF-Tiny-YOLO model integrating DenseNet for feature reuse to improve gradient propagation and detection accuracy.
• Implemented Resize and Re-organization (Reorg) with convolution kernel compression to reduce model parameters and increase detection speed.
• Utilized 1x1, 1x1, and 3x3 convolution kernels at the network terminal for feature dimensionality reduction and increased network depth without computational overhead.

Main Results:

• Achieved a mean average precision (mAP) of 99.99% and an average intersection over union (IoU) of 90.88%.
• Reached a detection speed of 280 frames per second (FPS).
• Demonstrated significantly improved detection performance compared to Tiny-YOLO and YOLOv2 models.

Conclusions:

• The DF-Tiny-YOLO model offers a fast and effective solution for detecting common apple leaf diseases.
• The proposed model advancements lead to superior accuracy and efficiency in automated disease identification.