An Crop-Gathering System Capable of Smart Image Recognition for Farmland
碩士 === 南臺科技大學 === 資訊工程系 === 107 === The paper proposes “An Crop-Gathering System Capable of Smart Image Recognition for Farmland”. Agriculture and judgment decisions because of requires long-term accumulation. However, we can recognize the model rules based on wisdom. Differentiating the maturity an...
| Main Authors: | , |
|---|---|
| Other Authors: | |
| Format: | Others |
| Language: | zh-TW |
| Published: |
2019
|
| Online Access: | http://ndltd.ncl.edu.tw/handle/sabq54 |
| Summary: | 碩士 === 南臺科技大學 === 資訊工程系 === 107 === The paper proposes “An Crop-Gathering System Capable of Smart Image Recognition for Farmland”. Agriculture and judgment decisions because of requires long-term accumulation. However, we can recognize the model rules based on wisdom. Differentiating the maturity and immature state of crops through neural network training. Finally, the crops are collected by a robotic arm. We propose to use “Multilayer Perceptron Machine Learning(MLP Machine Learning)” model to predict the position of a multi-axis robotic arm. Using the object detect the pixel coordinates of the center point of the crop on the image as the input of the neural network. Robotic arm as output.
From the object detect side, the paper proposes to use MobileNet version2.0 convolution neural network as a model for feature extraction. Feature extraction model combined with Single Shot MultiBox Detector(SSD) as object detection model for back-end layer of convolutional neural network. Crop detection through self-collection and labeling of images. Through experiments, it is also proved that the model training results proposed in this paper have an average accuracy of about 84%. Average accuracy is higher than other models. The average accuracy of the results of the robotic arm gathering is about 89%.
|
|---|