Design plan of crop disease detection system based on edge artificial intelligence
Source: InternetPublisher:通通 Keywords: artificial intelligence machine learning Updated: 2023/12/22
Crop disease damage is a major concern for farmers, and this project work uses machine learning to determine the types of diseases present in crops based on photos of leaves. First, use Single Shot Detector to detect leaves individually in real time from a video feed captured in the field. Images of leaves taken from the field can also be used as input to the system. A convolutional neural network is proposed to classify the types of diseases present in crops. The network is trained using the PlantVillage dataset. The proposed hybrid network is implemented on Kria KV-260 for real-time detection and identification. So this platform is developed for advanced vision application development and does not require complex hardware design knowledge. KV260 also offers the benefit of differentiating our designs at the software level through Vitis AI. The achieved disease classification accuracy is around 95.88%. To combat losses caused by crop diseases, edge AI systems based on machine learning can detect diseases and help farmers increase yields.
Motivation - Why did we decide to do this project?
Agriculture is extremely important to the world economy. Today, much of the crop vegetation fails because crop disease detection is not successfully integrated into farmers’ harvesting processes. Every year, farmers struggle with disease damage to their crops. Farmers can benefit greatly from early detection and treatment of these diseases. It is difficult to find skilled experts in this field who can detect any type of plant disease. It would be a boon to farmers if automated systems could identify crop diseases and other issues such as malnutrition, weeds or insect damage in real time via handheld devices or hardware on farm equipment.
Therefore, there is a need for a system that can predict crop diseases before the entire harvest is destroyed. Machine learning can be used to detect crop diseases and help farmers identify them. This research project uses the concept of deep learning to build a real-time plant disease detection system. The model can be deployed on embedded platforms such as Kria KV260 to detect the presence of diseases in crops in real time. The main purpose is to effectively predict plant diseases so farmers can take effective measures before the disease spreads to crops.
Therefore, the goals of this project are as follows:
Collection of crop disease data set for Indian crops (Gujarat)
Develop machine learning (ML) models for crop disease detection and classification.
Porting ML models on Kria KV 260
The main goal of this project is to efficiently detect leaves on plants and then accurately identify the type of disease present on the leaves. The SSD model is used to identify plant leaves, and a new architecture based on convolutional neural network (CNN) is used to identify leaf diseases. The SSD model and the proposed CNN model are combined to create a hybrid model that can detect leaves and diagnose diseases simultaneously. Furthermore, the proposed hybrid model was deployed on Kria KV-260 for real-time testing to solve the problem of real-time detection of plant leaf diseases. The figure below depicts the block diagram of the proposed system for leaf identification and disease classification.
result:
The proposed system was tested on leaf images from the PlantVillage dataset as well as data captured from a nearby real tomato farm. The effectiveness of the model in leaf detection and disease identification was tested by applying the system to disease-infected tomato leaves. The result is shown below:
As can be seen from the figure, the model is able to accurately identify the types of diseases - spider mite, early blight, tomato mosaic virus and leaf mold - from the leaves. As demonstrated by these real-time field tests, the proposed model performs well under all conditions, including atmosphere, background, soil, and lighting.
Problems you may face:
1. Install Vitis on Ubuntu
Installing Vitis and Vivado on Ubuntu is very tiring. They are prerequisites for installing Vitis and require a large number of dependencies. There is no such proper documentation or link available for proper installation. Additionally, it takes a long time to install into the device. After referring to some dead links, it took us over 12 hours to complete the installation.
2. TensorFlow Frozen graph problem and its installation
To obtain the final static graph, the inputs are .pb and .ckpt files, which gives us the output frozen_graph.pb. This is a major issue with the TensorFlow library, without this graph no further processing is possible. In the end this instruction was not executed:
freeze_graph --input_graph yolov2-tiny.pb --input_checkpoint yolov2-tiny.ckpt --output_graph freeze/frozen_graph.pb --output_node_names yolov2-tinyconvolutional9/BiasAdd --input_binary true
TensorFlow installation issues
3.OpenCV error
For real-time interfacing and processing of camera modules, open CV is the most widely used Python library. Without this, the camera interface does not happen, and installing it in a Linux environment is very time-consuming.
4. Install Vitis AI
To install Vitis AI, we tried to git clone the KV-260 ml acceleration library, but the clone got stuck at some point due to some issues. For installation purposes, we need to create a Docker for installing Vitis-AI. There are bugs related to index packages, GnuTLS, and early EOF.
5. Implementation of real-time webcam interface during YOLO v2-v3 on Kria Kv-260
Before starting with our own model, we considered using YOLO to implement and check the hardware. However, the main problem with live camera interfacing with the KV-260 when implementing the YOLO pre-trained model is that the kit does not support live webcam interfacing.
After summarizing the problems we may encounter, the project is over.
- Battery equalization system and control method thereof
- Ways to deliver more than 100A to ADAS processors
- What components and functions does the battery management system consist of?
- How to build a self-driving drone
- Ray detection circuit
- Computer host overheating alarm circuit
- Two-channel SL320 level indication circuit
- SCR control buzzer circuit
- Circuit diagram of Shenlong-Fukang DC7140 car diagram reading method example
- One use and one standby constant pressure water supply frequency conversion speed control circuit b