CN7023 Artificial Intelligence and Machine Vision Assignment Sample
FRUITS CLASSIFICATION USING CONVOLUTIONAL NEURAL NETWORK
Abstract:
The fruits classification using the Convolutional Neural Network is the one of the important applications in the computer vision. In this process using the classification to recognize the fruits based on the layer and images. Still over there is some problem to recognize the fruit based on the weighting scale due to the complexity and similarity of the fruit. In this paper the fruit classification using Convolutional Neural Network is proposed and using the deep classification techniques for the process. here used the data set Fruits – 360 for the analysis process in that it contains the 67692 images from the 131 different classes. First the image is dividing into the training and testing process. here used various combination of hidden layer for improve the classification accuracy of the images in the different cases. Analysed various data losses during the process and obtained the best accuracy of testing process 100% and training accuracy is 99.32%.
-
Introduction
In the last few years, the human pays more attention in the food we eat. In specifically the computer visions mostly used the fruit recognition method. This is done using the field of the image classification and recognition methodology. The Deep Neural Network is used for the predict the fruits from the collection of images and process with the further research. Compare with the machine learning algorithm the Deep Neural Network (DNN) is performed better and process with the efficient manner. In the Deep learning process, the Convolutional Neural Network (CNN) is used as the one of type of Artificial Neural Network (ANN). This technology is used for the different kind of recognition of the images that is include with the video and image, handwritten recognition and face recognition and fruit recognition like that. The accuracy of the fruit recognition by using CNN is very similar to human recognition process. The Convolutional Neural Network is having the similar architecture of the Artificial Neural Network that is consist of the pooling layers along with the extraction and convolution and also having the combine process of high-level process from the 2D input. Like that it is used in the different kind of purposes.
The research questions are based on following manner.
- How to apply CNN in the fruit image?
- What are factors considered for the measuring the accuracy level?
- How select the better algorithm for this analysis?
In this research implemented the fruit recognition classifier using the Convolutional Neural Network and input image is considered with the RGB image and used the various combination of hidden layer to obtain the best performance of the network and analysis the accuracy level. The final result note and the implementation of this process is done using python.
-
Methodology
The process of accurate and efficient fruit recognition method is established with the help of the Neural Network. It is very important in the term of robotic harvesting according to the different functionality. The fruit recognition system is deal with the accuracy and implement with the trained data set that is take it for the process. it is showing the real time prediction of the various types of fruits according to their various properties. Based on the classifier (C. Hung, 2020) recognition of fruits are considered with the many number of factors that is considered with difficult based on the process and measurement of accuracy level of the process (J. Hemming, 2018). The factors are considered like scenes of fluctuating brightness, sharpen edge of the images, the texture and reflection properties of the images and process by other objects like that it is considered with the different kind of process and analysis.
The main problem in the fruit recognition is considered with the image segmentation problem related to the different process and measurement. There are several problems area addressed by the various authors. (A. Krizhevsky, 2018) Establish the system that is able to detect the apples based on the colour. It is related to the survey if the yield prediction and apple detection of the process. (F. Siddique, 2019) author is proposed the five-layer level of the segmentation method to improve the prediction level of the different kind of measurement-based analysis according to the different kind of attributes based on the selected factor. This method is used with the Sparse Autoencode (SAE) based on their features mapping with the CRF framework according to the different functionality and mapping analysis.
The Novel approach is used for the detecting the fruit using the Convolutional Neural Network (A. Krizhevsky, 2018) along with the Faster Region based learning system. This is used for the image prediction done by the deep analysis according to the different kind of analysis. In this they trained the model using the RGB and Near infrared images for easy predicting and processing of easy analysis and process of the analysis and process of the measurement according to the specification of different system and movement for effective analysis. The combination RGB and NIR is provide the effective mapping of the images and also predict the starting and late fusion of the images based on the technology. The fruit detection is achieved highly based on the remarkable system according to the easy prediction of the method that is able to process different kind of analysis system (J. Hemming, 2018) . Like that different kind of methodology is analysed and used in the measurement.
-
Simulations
The fruit classification is done based on the selected data set. That is used for this analysis and provide the effective result towards the process and measurement.
3.1 Data set
The dataset is considered for this analysis is Fruits – 360 for the analysis process in that it contains the 67692 images from the 131 different classes. First the image is dividing into the training and testing process.
Figure 1: Loading images of train and test data
This is shows how the dataset is considered and loaded into the proper link for the analysis using the Python. Here the fruits – 360 is considered in this the unstructured data is also considered for the effective analysis and prediction process.
Figure 2: Training Dataset and Test data Images
The above image shows the result of after training and testing the data set according to the different kind of analysis based on the terminology of easy access based on the terminology of process. In that found the 67692 images based on the 131 classes according to different measurement and process analysis.
3.2 Encode the image into the Neural Network
In this considered the total number of images: 90483. And the training set size: 67692 images (one fruit or vegetable per image). Then considered with the test set size: 22688 images (one fruit or vegetable per image). Based on that the number of Number of classes: 131 (fruits and vegetables). The image size considered for this classification is done based on the fruits is 100×100 pixels. Then considered with the filename format: imageindex100.jpg (e.g. 32100.jpg) or rimageindex100.jpg (e.g. r32100.jpg) or r2imageindex100.jpg or r3imageindex100.jpg. “r” stands for rotated fruit. “r2” means that the fruit was rotated around the 3rd axis. “100” comes from image size (100×100 pixels). Then considered with the different varieties of the same fruit (apple for instance) are stored as belonging to different classes (Y. LeCun, 2019).
3.3 Network Architecture
The network architecture is considered for this analysis is based on the different kind layer and measured the different kind of attributes based on the value of each process. the two convolutional layers is considered and this is followed by the polling layers. That two layers are connected to each other for the easy communication and process measurement of attributes (K. Fukushima and S. Miyake, 2019). The first process the network layer is considered with the input of the 30,000 neurons that is represented with the image size of 100X100based on the RGB size. The first layer is considered as the convolutional layer 1 that is having the 64 filters along with the kernel functional based on the 3X3 image size of pixels and the rectified linear is the value of the measurement based on the terminal value according to the specific function and measurement of each process. The second layer is considered as the convolutional layer 2 that is having the 64 filters along with the kernel functional based on the 3X3 image size of pixels and the rectified linear is the value of the measurement based on the terminal value according to the specific function and measurement of each process. in the ReLU is placed as the first layer of the process and then it is carry over with the specified value (J. Hemming, 2018). This technology is used for the different kind of recognition of the images that is include with the video and image, handwritten recognition and face recognition and fruit recognition like that. The accuracy of the fruit recognition by using CNN is very similar to human recognition process. The Convolutional Neural Network is having the similar architecture of the Artificial Neural Network that is consist of the pooling layers along with the extraction and convolution along with the different kind of analysis and process of the measurement based on the terminology of access and process. like that the architecture is created and executed with the different kind of analysis and process.
The convolutional layer is sued for the feature extraction of the input data according to the specific access and measurement of the functionality and process analysis of the measurement. Like that it is process for the efficient prediction using the layer-by-layer process according to the CNN (Y. Sun, 2020). The kernel initializer is representing with the filtering process according to the determination of the layers. The ReLU is used for the activating the process according to each access and measurement of the functionality. Then it is enhancing the functional of the convolutional layer based on the terminology. The communication of the layer 1 and layer 2 is considered with the effective process and measurement of the technology according to the specification of image attributes based on that it is executed for the different analysis.
The regularization layer is established with the good process and measurement of the technology according to the different factors and process. it is having the probability of the 0.25% along with the pooling layer 2 that is accessing and randomly switching by the different kind of process and analysis of each process. like that it is measured and apply into the different kind of images. The neurons in the images are during the training process it is reduce the overfitting based in the attribute and also improve the efficiency of the process by applying the performance of the network (D. H. Hubel and T. N. Wiesel, 106-154). It is used for the network to suitable for the efficiency generalization and less compelling to the overfit of the process according to the functionality of each analysis. The flatter layer is used for the analysis of the convert the 2D filter matrix into the various ID of feature vector according to the functionality connected layers.
-
Result obtained
Using the above methodology CNN, the following results are obtained. First the CNN model is constructed in the following manner that is noted in the below process. (Figure 4). It is considered with the layer and convolutional model of the different kind of analysis along with the spatial drop out and pooling layer value based on that image accuracy will be designed and implemented with the various analysis of the process according to the different process and measurement. The spatial value is changed based on the image selection and process of the method that is executed efficiently of the measurement.
4.1 Calculating Loss and accuracy
Next considered with the calculation of the loss and accuracy of the image based on each pixel value of the image. The above results is shows that the testing loss is happen in the measurement of 0.023% only. It is considered nearly to the 0% and the test accuracy is measured with the value of 99.32%. That is nearly 99% like that the accuracy and test loss is calculated using the above process and measurement of the value based on each measurement of the image. This is shows that the CNN Classification is executed properly and measure with the proper execution of each image classification according to the different kind of factors. If the image classification accuracy is increased means then it is mapped with the specified technology and suitable process. Here fruit classification is done with the accuracy of 99% along with the 0% loss of the images using the fruits 300 dataset. Based on the CNN is suitable for this kind of fruit classification based on the terminology and measurement of each access and process (C. Hung, 2020).
-
Critical Analysis of Result
Based on the above process we got the result of fruit classification is done with the accuracy of 99% along with the 0% loss of the images using the fruits 300 dataset. For this classification is done using the Convolutional Neural Network process (CNN). It is considered with the layer and convolutional model of the different kind of analysis along with the spatial drop out and pooling layer value based on that image accuracy will be designed and implemented with the various analysis. The plotting of the loss and accuracy is following manner.
For this analysis considered the data set of Training set size 67692 images that is having one fruit or vegetable per image. And considered the test set size 22688 images that is having one fruit or vegetable per image. Based on that the different varieties of the same fruit apple for instance are stored as belonging to different classes. Like that the fruit classification is done based on the spatial drop out and maximum pooling of the process according to the different functionality and measurement of the process this is used in the different kind of image classification based on the requirement of the process (A. Krizhevsky, 2018). The kernel initializer is representing with the filtering process according to the determination of the layers. The fruit recognition classifier using the Convolutional Neural Network and input image is considered with the RGB image and used the various combination of hidden layer to obtain the best performance of the network and analysis the accuracy level. The final result note and the implementation of this process is done using python.
The ReLU is considered for the activating various pooling process according to the access of different process and analysis of measurement. In the plot is represent the accuracy level is gradually increased and the loss is gradually decreased based on that it is showing the fruit classification is done through the different process and measurement of the technology.
-
Conclusion
In this research is fully based on the fruit classification using the convolutional Neural Network (CNN). For this analysis the data set considered is Fruits – 360 for the analysis process in that it contains the 67692 images from the 131 different classes. The fruit recognition classifier using the Convolutional Neural Network and input image is considered with the RGB image and used the various combination of hidden layer to obtain the best performance of the network and analysis the accuracy level. The final result note and the implementation of this process is done using python. The image is dividing into the training and testing process. Here used various combination of hidden layer for improve the classification accuracy of the images in the different cases. Here fruit classification is done with the accuracy of 99% along with the 0% loss of the images using the fruits 300 dataset using the CNN. Like the fruit classification is done using the CNN is implement using python and analysis the accuracy and losing level of the image classification process.
Bibliography
- Karpathy, G. T. S. S. T. L. R. S. F.-F., 2016. “Large-scale video classification with convolutional neural networks. in Proceedings of the IEEE conference on Computer Vision and Pattern Recognition, pp. 1725-1732..
- Krizhevsky, I. S. a. G. E. H., 2018. Imagenet classification with deep convolutional neural networks. Advances in neural information processing systems,, pp. 1097-1105..
- Hung, J. N. Z. T. J. U. a. S. S., 2020. “Orchard fruit segmentation using multi-spectral feature learning. in 2013 IEEE/RSJ International Conference in Intelligent Robots and Systems, pp. 5314-5320..
- H. Hubel and T. N. Wiesel, 106-154. “Receptive fields, binocular interaction and functional architecture in the cat’s visual cortex,”. The Journal of physiology, p. 2019.
- Siddique, S. S. a. M. A. B. S., 2019. Handwritten Digit Recognition using Convolutional Neural Network in Python with Tensorflow and Observe the Variation of Accuracies for Various Hidden Layers. Sensors, , pp. 2- 9.
- Cheng, L. D. Y. S. a. M. B., 2019. Early yield prediction using image analysis of apple fruit and tree canopy features with neural networks. Journal of Imaging, pp. 5 – 9.
- Sa, Z. G. F. D. B. U. T. P. a. C. M., 2019. Deepfruits: A fruit detection system using deep neural netowrks. Sensors, , pp. 1222-1234.
- Hemming, J. R. J. H. a. E. v. H., 2018. Fruit detectability analysis for different camera positions in sweetpepper. Sensors, , pp. 6032-6044.
- Fukushima and S. Miyake, 2019. Neocognitron: A self-organizing neural network model for a mechanism of visual pattern recognition,”. in Competition and cooperation in neural nets Springer, , pp. 745-758..
- Schaul, S. Z. a. Y. L., 2017. “No more pesky learningrates. in International Conference on Machine Learning, pp. 343-351..
- LeCun, B. B. J. S. D. D. H. R. E. H. H. e. a., 2019. Backpropagation applied to handwritten zip code recognition,. Neural computation, pp. 541-551,.
- Sun, Y. C. X. W. a. X. T., 2020. Deep learning face representation by joint identification-verification. in Advances in neural information processing systems, pp. 1988-1996..
Know more about UniqueSubmission’s other writing services:
