Annotation-free Quality Estimation of Food Grains using Deep Neural Network

We propose a fast and accurate system for automatically estimating the quality of food grains on resource constrained portable devices using computer vision. We are motivated by an urgent need in India for grain quality estimation to ensure transparency in the agricultural supply chain and empower poor farmers to get the correct price for their crops. The system uses instance segmentation of touching grains, followed by classification of each grain according to E-NAM1 parameters. To the best of our knowledge, this is the first attempt to use Deep Learning to estimate quality of cluttered sample of grains using only mobile phone. Samples are collected from various Agricultural Produce Market Committee (APMC) yards, which are used to generate synthetic data to simulate realistic clutter of grains for training our instance-segmentation network. Novel augmentation techniques while training make the system robust to illumination changes. Our system obtains the state-of-the-art performance and has been tested in various locations in India. At a mAP score of 0.74 and classification accuracy 92%, our system takes less than 100s as compared to 15 minutes of manual quality estimation.

[1]  Kaiming He,et al.  Mask R-CNN , 2017, 2017 IEEE International Conference on Computer Vision (ICCV).

[2]  Peyman Moghadam,et al.  Deep Leaf Segmentation Using Synthetic Data , 2018, BMVC.

[3]  Ing Ren Tsang,et al.  Multiclass Weighted Loss for Instance Segmentation of Cluttered Cells , 2018, 2018 25th IEEE International Conference on Image Processing (ICIP).

[4]  Alexey Shvets,et al.  TernausNetV2: Fully Convolutional Network for Instance Segmentation , 2018, 2018 IEEE/CVF Conference on Computer Vision and Pattern Recognition Workshops (CVPRW).

[5]  Alexei A. Efros,et al.  Image-to-Image Translation with Conditional Adversarial Networks , 2016, 2017 IEEE Conference on Computer Vision and Pattern Recognition (CVPR).

[6]  T. Pearson,et al.  High-Speed Sorting of Grains by Color and Surface Texture , 2010 .

[7]  Forrest N. Iandola,et al.  SqueezeNet: AlexNet-level accuracy with 50x fewer parameters and <1MB model size , 2016, ArXiv.

[8]  Pekka Ruusuvuori,et al.  Computational Framework for Simulating Fluorescence Microscope Images With Cell Populations , 2007, IEEE Transactions on Medical Imaging.

[9]  Cewu Lu,et al.  Annotation-Free and One-Shot Learning for Instance Segmentation of Homogeneous Object Clusters , 2018, IJCAI.

[10]  Kadir Sabanci,et al.  Grain classifier with computer vision using adaptive neuro-fuzzy inference system. , 2017, Journal of the science of food and agriculture.

[11]  Luis Perez,et al.  The Effectiveness of Data Augmentation in Image Classification using Deep Learning , 2017, ArXiv.

[12]  Vijay Vasudevan,et al.  Learning Transferable Architectures for Scalable Image Recognition , 2017, 2018 IEEE/CVF Conference on Computer Vision and Pattern Recognition.

[13]  Thomas Brox,et al.  U-Net: Convolutional Networks for Biomedical Image Segmentation , 2015, MICCAI.

[14]  Cewu Lu,et al.  SRDA: Generating Instance Segmentation Annotation via Scanning, Reasoning and Domain Adaptation , 2018, ECCV.

[15]  David Svoboda,et al.  TRAgen: A Tool for Generation of Synthetic Time-Lapse Image Sequences of Living Cells , 2015, ICIAP.

[16]  Yoshua Bengio,et al.  How transferable are features in deep neural networks? , 2014, NIPS.

[17]  Mark Sandler,et al.  MobileNetV2: Inverted Residuals and Linear Bottlenecks , 2018, 2018 IEEE/CVF Conference on Computer Vision and Pattern Recognition.

[18]  Jian Sun,et al.  Deep Residual Learning for Image Recognition , 2015, 2016 IEEE Conference on Computer Vision and Pattern Recognition (CVPR).

[19]  Shu Liu,et al.  Path Aggregation Network for Instance Segmentation , 2018, 2018 IEEE/CVF Conference on Computer Vision and Pattern Recognition.