A novel image measurement algorithm for common mushroom caps based on convolutional neural network

Abstract In a mushroom greenhouse, for example, there are no sensors that can measure the diameter of clustered mushrooms, and farmers can only empirically estimate the relationship between the mushroom growth and greenhouse microclimate data. Our team developed an image measurement system that can record the circle diameter of Agaricus bisporus (hereinafter referred to as “common mushroom”) caps during the fruiting period. To continuously record the size of the mushroom caps for purpose-specific analysis (such as data analysis for optimized greenhouse microclimate control, calculation of growth rate, or harvest reminder), a new algorithm is proposed for calculating the diameter of the circle of round mushroom caps. This algorithm is used to estimate the mushroom circles using the images continuously captured by a camera and generate a growth record of the mushroom caps. The automatic adjustment of the camera aperture and focus during the growth of the mushroom results in a color deviation between images, which makes it difficult to detect the circles of the mushroom caps. The proposed algorithm is sufficiently robust to overcome the challenges presented by the color deviation in achieving accurate estimation. It is based on the identification results of the convolutional neural network and combines the innovative Score-Punishment (SP) algorithm in this study to calculate the circle diameter of the mushroom caps. The proposed algorithm outperformed the current Circle Hough Transform (OpenCV’s implementation). Besides, the proposed algorithm is more robust without the need to set parameters for the images of each time period, making it more feasible for practical application. In addition to describing the algorithm in detail, in this work, the effectiveness and practicality of the proposed algorithm through experiments were also verified.

[1]  R. Noble,et al.  AE—Automation and Emerging Technologies , 2001 .

[2]  C. Igathinathane,et al.  Sunflower floral dimension measurements using digital image processing , 2018, Comput. Electron. Agric..

[3]  Xunmu Zhu,et al.  Automatic recognition of lactating sow postures from depth images by deep learning detector , 2018, Comput. Electron. Agric..

[4]  Jiun-Jian Liaw,et al.  Development of a Mushroom Growth Measurement System Applying Deep Learning for Image Recognition , 2019, Agronomy.

[5]  Kesheng Wu,et al.  Fast connected-component labeling , 2009, Pattern Recognit..

[6]  Deqin Xiao,et al.  Feeding behavior recognition for group-housed pigs with the Faster R-CNN , 2018, Comput. Electron. Agric..

[7]  S. J. Miles,et al.  Influence of Mushroom Strains and Population Density on the Performance of a Robotic Harvester , 1997 .

[8]  J. Kittler,et al.  Comparative study of Hough Transform methods for circle finding , 1990, Image Vis. Comput..

[9]  Yongsheng Si,et al.  High-Throughput Phenotyping of Seed/Seedling Evaluation Using Digital Image Analysis , 2018 .

[10]  Andrey Somov,et al.  Pervasive Agriculture: IoT-Enabled Greenhouse for Plant Growth Control , 2018, IEEE Pervasive Computing.

[11]  Geoffrey E. Hinton,et al.  ImageNet classification with deep convolutional neural networks , 2012, Commun. ACM.

[12]  Dong Wang,et al.  Dairy goat detection based on Faster R-CNN from surveillance video , 2018, Comput. Electron. Agric..

[13]  Zheng Yuan,et al.  Image-Based On-Panicle Rice [Oryza sativa L.] Grain Counting with a Prior Edge Wavelet Correction Model , 2018, Agronomy.

[14]  Avital Bechar,et al.  Robotic Disease Detection in Greenhouses: Combined Detection of Powdery Mildew and Tomato Spotted Wilt Virus , 2016, IEEE Robotics and Automation Letters.

[15]  Geoffrey E. Hinton,et al.  Deep Learning , 2015, Nature.

[16]  Kaiming He,et al.  Faster R-CNN: Towards Real-Time Object Detection with Region Proposal Networks , 2015, IEEE Transactions on Pattern Analysis and Machine Intelligence.