Best way to train a model in keras r [closed] - r

Question

Closed. This question needs to be more focused. It is not currently accepting answers.
Want to improve this question? Update the question so it focuses on one problem only by editing this post.
Closed 4 years ago.
Improve this question
Overview
I'm not new to R but am very new to machine learning.
For work I collect data by writing on a datasheet printed on waterproof paper which I then have to transcribe to the database manually. This takes a long time at the end of a long day and is a process prone to mistakes.
The entire datasheet is shown below
What I would like to do is simply take a photo of the sheet and have keras read it and input the results into a database
And the section of the datasheet that I am interested in getting Keras to read is shown here
Each row of the datasheet represents what species of coral was found and each column represents what transect it was found on ie 7 Acroppora was found on T1
Each of these cells are given a unique entry in the database in a format similar to this which would show how the Acropora row is recorded
For each datasheet that we have entered in the past (probably somewhere between 1000 and 2500) there are corresponding database entries which can be exported to csv and linked to each datasheet
Ultimately, what I would like to do is simply take a photo of the sheet and have keras read the part I'm interested in (shown in second image) and input the results into a CSV in a similar format shown in the third image
The questions
What I've been thinking about is getting it to identify the borders of the parts of the datasheet I'm interested in (shown in the second image) and extract it. This would mean that I could then put in coordinates for each cell, ie Acropora T1 (as shown in the image below) and identify the number counted in that cell and export it to a database
Does this process sound possible? If so, would anyone know of any examples I could look up or even what you would call this process so I can look it up
Otherwise I was thinking about scanning each sheet as a whole (As shown in the first image) and simply training from that, however I feel that would be more prone to errors
I really hope this makes sense and would very much appreciate any help and/or suggestions either specifically to the questions that I asked or about my project in general

Answer 1

This uses OpenCV and Python.
According to the chapter on 'Hough Line Transform' you could detect lines like this.
import cv2
import numpy as np
img = cv2.imread('D:/Books/lines1.jpg', cv2.IMREAD_GRAYSCALE)
edges = cv2.Canny(img,50,150,apertureSize = 3)
cv2.imwrite('D:/Books/edges.jpg',edges)
But based on my simple research I think counting is possible using code like this.
More knowledge of OpenCV is required at this stage. I think this is just dilating and the borders of the lines are more pronounced.
img = cv2.imread('D:/Books/lines1.jpg', cv2.IMREAD_GRAYSCALE)
edges = cv2.Canny(img,50,150,apertureSize = 3)
cv2.imwrite('D:/Books/edges.jpg',edges)
kernel = cv2.getStructuringElement(cv2.MORPH_ELLIPSE, (4, 4))
dilated_Edges = cv2.dilate(edges, kernel, iterations=1)
cv2.imwrite("D:/Books/dilated_Edges.jpg", dilated_Edges);
lines = cv2.HoughLines(image=dilated_Edges,rho=1,theta=np.pi/180, threshold=100)
print( len(lines))
This prints 8 for me which isn't correct.
I pursued this and this code is based on help from the OpenCV forum(Suleyman TURKMEN).
Images I tested with are these. Prints the correct count.
import cv2
import math
img = cv2.imread('D:/Books/lines1.jpg', cv2.IMREAD_GRAYSCALE)
ret,bw = cv2.threshold(img,0,255,cv2.THRESH_BINARY_INV + cv2.THRESH_OTSU)
cv2.imshow("bw", bw)
kernel = cv2.getStructuringElement(cv2.MORPH_ELLIPSE, (5, 2))
eroded_Edges = cv2.erode(bw, kernel, iterations=3)
dilated_Edges = cv2.dilate(eroded_Edges, kernel, iterations=4)
im2, contours, hierarchy = cv2.findContours(dilated_Edges, cv2.RETR_LIST, cv2.CHAIN_APPROX_SIMPLE)
print (len(contours) , " horizontal lines")
cv2.imshow("vertical lines", eroded_Edges)
kernel = cv2.getStructuringElement(cv2.MORPH_ELLIPSE, (1, 5))
eroded_Edges = cv2.erode(bw, kernel, iterations=3)
im2, contours, hierarchy = cv2.findContours(eroded_Edges, cv2.RETR_LIST, cv2.CHAIN_APPROX_SIMPLE)
print (len(contours) , " vertical lines")
cv2.imshow("horizontal lines", eroded_Edges)
cv2.waitKey()