Automatic Quantification of wheat spikes using Neural Networks

Project Title

Automatic Quantification of wheat spikes using Neural Networks

Project Area of Specialization Computer ScienceProject Summary

Crop Yield is an essential measure for breeders, researchers, and farmers and is composed of and may be calculated by the number of ears per square meter, grains per ear, and thousand grain weight. Manual wheat ear counting, required in breeding program to evaluate crop yield potential, is labor-intensive and expensive; thus, the development of real-time wheat head counting system would be a significant advancement in the field of agricultural and farming sector.

In this Project idea we would like to do data collection by take picture sample of about 500 wheat spikes and make a dataset and using segmentation an image into superpixels by using an algorithm technique Simple Linear Iterative clustering ( SLIC ), and them make a deep convolutional neural network for semantic segmentation of wheat spikes. As the method is proposed to be a deep learning model so required a features of machine learning.

This Project can be very helpful in field of agriculture and farming sectors, which can be free from human error and can be done automatically through by training our machine with the neural network regarding the effect of climate changes.

As this project can be best helpful for future use and according to my study it will be easy for me to work on it. Looking to the positive aspects of the project I want you to allow me and propose that project, so I can start doing my work on it.

Project Objectives

Main objective of this project is to develop a system which will help us in counting of wheat spikes. Using this system farmers and agricultural dept sector, can easily quantify yearly yield, number of ears per square meter, Grains per ear and quality of the yield (wheat yield). For this purpose, I will develop an android/IOS app which farmers will use to quantify their data by providing some basic credentials, which will be processed with our data and provide them with required results. Using Convolutional neural networks, I will train my model, Which will be used during the Quantification process. 

Project Implementation Method

The high quality in-feld images from this field trial are used to construct the SPIKE data set. In which about 300 images of ten wheat varieties are taken at different growth stages using a high resolution camera. Then the images are automatically cropped so the sides areas have to be ignored and a specific image has been taken, next the the images are manually annotated with bounding boxes highlighting all the spikes present in the images. The images and annotations are then feed to the convolutional Neural Networks (CNN) for training.

Using Video Object Tagging Tool provided by Microsoft we do annotation of images and do image labeling using this publicaly available Microsoft tool.

Model Development :

The SPIKE data set of 335 images in total was split into 305 training images and 30 testing images. This split was performed at the image level, not at the spike level, to ensure that no spikes from the same image could be seen in both training and testing sets.
 

R-CNN Model :

Region-based Convolutional Neural Network (R-CNN) was introduced by Girshick et al. for object detection using a selective search to detect regions of interest and a CNN to classify them. Later, Fast R-CNN by ROI pooling was used after final convolution to extract a fixed length feature vector from the feature map along with the training of all network weights with back-propagation.
Later, Faster R-CNN was developed by Ren et al. This model consists of two networks: a region proposal network (RPN) for generating region proposals, and a convolutional network which takes the proposed regions to detect objects almost in real-time. The main difference between the two region-based methods is that, to generate region proposals, Fast R-CNN uses selective search whereas Faster R-CNN uses high-speed RPN and shares the bulk of the computation time with object detection.
Briefly, RPN ranks the region boxes (called anchors) and proposes the ones that are most likely to contain the desired objects. Due to its fast processing capability and high recognition rate.

This train model will then be used in andorid app, which is to be implement later after training the model and connect that trained model with android app using APIs.

Benefits of the Project

This Project can be very helpful in field of agriculture and farming sectors, which can be free from human error and can be done automatically through by training our machine with the neural network regarding the effect of climate changes.

Wheat is one of the most globally significant crop species with an annual worldwide grain production of 700 million tonnes . In recent years, however, there is an increasin demand for grain. At the same time, the seasonal fluctuations, the extreme weather events and the altering climate in various regions of the world, increase the risk of inconsistent supply. This points to the need to identify hardier and higher yielding plant varieties to both increase crop production and improve plant tolerance to biotic and abiotic stresses. To discover higher-yielding and more stress tolerant varieties, biologists and breeders rely more and more on high-throughput phenotyping techniques to measure various plant traits, which in turn are used to understand plant’s response to various environmental conditions and treatments, with the hope to improve grain yield.

Estimating the yield of cereal crops grown in the field is a challenging task, yet it is an essential focus of plant breeders for wheat variety selection and improved crop productivity. In this we have presented the first deep learning models for spike detection, trained on wheat images taken in the field. The models are capable of accurately detecting wheat spikes within a complex and changing imaging environment.

As this project can be best helpful for future use and according to my study it will be easy for me to work on it. Looking to the positive aspects of the project I want you to allow me and propose that project, so I can start doing my work on it.

Technical Details of Final Deliverable

This train model will then be used in android app, which is to be implement later after training the model and connect that trained model with android app using APIs.

So the final deliveriable will be a mobile app in which every users for-example: a farmer will take a picture of any field and then add then to the mobile app using choose file option after then the pictures will be passes from the neural network which is connected with the mobile using an APIs. After this the user will be provided with the spikes counting details and also with the estimate weight of wheat spikes.

Final Deliverable of the Project Software SystemCore Industry ITOther Industries Agriculture , Food Core Technology Artificial Intelligence(AI)Other Technologies OthersSustainable Development Goals Decent Work and Economic GrowthRequired Resources

Item Name	Type	No. of Units	Per Unit Cost (in Rs)	Total (in Rs)
			Total in (Rs)	54000
Camera	Equipment	1	40000	40000
The ground-based vehicle for imaging in the field	Equipment	1	5000	5000
Drone for	Equipment	1	9000	9000