Abstract:
Here are some problems due to which we proposed new system to overcome from the
following problems; the SVM based Fruit Quality System just not detect the fruits of quality
rapidly. The system takes so much time for displaying the quality because of different sizes,
which is very consuming of time. In it the inspection of fruit does not perform accurately. In
this System, the area of the fruit which is covered, the ratio, perimeter and the centroid was
not used and displays the quality of fruit and nothing else and it is expensive computationally
and execute slow. In SVM, for the calculation of error rate of fruit is calculated by the mean
of square error and neighbor error rate. But in K-NN algorithm you just have to take some
image then it computes the total error rate which is very easy method as compare to SVM.
The thesis statement of this project is to identify the fruit and the defects in fruits. Several
techniques of image processing include in the project which consists of Features Extraction,
Edge detection and its techniques, Noise Filtering, Binary to Grey Scale Conversion, Image
Histogram, Shaipen an Image, Color Clustering,. This all are applied on the software which is
called “MATLAB” which is a research platform.
In this project used Machine Learning Algorithm (Supervised Learning) and K-Nearest
Neighbor (K-NN) for developing software. The benefit of this algorithm that it detects not
only individual fruit but also multiple fruit and produce result in a less amount of time and
accurately. The K-Nearest Neighbor Classifier (K-NN) applies on the training of fruits and
testing of fruits.
For training and testing of fruit, we gathered a large amount of data set in which different
fruits are involved.All the images are saved in their folder respectively. For instance if any
fruit is free from any folder then it place on non-defected folder and if any fruit are defected
then it placed on the defected folder respectively after training phase when complete.The
result which we obtained through K-Nearest Neighbor (K-NN) classifier is about 80 percent
as a quality rate, while 20 percent considered as an error rate which can occur in the model.
