You are on page 1of 3

STORE ACTIVITY RECOGNITION BASED ON

RASPBERRY PI

1st Kahlil Muchtar* 2nd Muhammad Jurej Al Hamdi 3rd Wildan Al Mumtaz
Department of Electrical and Computer Department of Electrical and Computer Department of Electrical and Computer
Engineering, Universitas Syiah Kuala Engineering, Universitas Syiah Kuala Engineering, Universitas Syiah Kuala
Telematics Research Center (TRC) Aceh, Indonesia Telematics Research Center (TRC)
Universitas Syiah Kuala chairuman@mhs.unsyiah.ac.id Aceh, Indonesia
Aceh, Indonesia mayafitria@unsyiah.ac.id
kahlil@usk.ac.id (corr.author)

4th Nizam Albar 5th Teuku Abrar Mifta 6th Jamalul Ridha
Department of Electrical and Computer Department of Electrical and Computer Department of Electrical and Computer
Engineering, Universitas Syiah Kuala Engineering, Universitas Syiah Kuala Engineering, Universitas Syiah Kuala
Aceh, Indonesia Aceh, Indonesia Aceh, Indonesia

Abstract— Tomatoes are a horticulture product with based optimization framework to detect recognized shop
the potential to be developed since they have a high activity..
economic value and are in high demand by industry and
II. RELATED WORK
consumers. Tomato plants, on the other hand, still need
to be handled carefully in order to boost yield harvest. Previous research by Mark Sandler, et al. [1] discusses
Furthermore, susceptibility of the tomato plants to the advanced performance of mobile models on many tasks
Septoria leaf spot disease, which arises due to Septoria and benchmarks and across a spectrum of different models
Lycopersici Speg fungal infection, is being one of the size. We also describe an efficient way to implement
challenges in escalating tomato production itself. this mobile model to object detection in a new framework
Regarding the problem, this study aims to detect we call it SSDLite. But as described in [2-3], they
Septoria leaf spot on tomato plants by developing a tool demonstrated how to build a mobile semantic segmentation
utilizing deep learning and Convolutional Neural model via a reduced form of DeepLabv3 which we call
Network (ConvNets or CNN) model. CNN model Mobile
conducted in this work is a supervised learning DeepLabv3. In [4], the authors discuss based on the
technique that extensively operated for solving linear inverse residual structure where shortcut connections exist
and non-linear problems. Moreover, Raspberry Pi between the thin layers of the bottle neck. Intermediate
microcontroller and Intel Movidius Neural Computing expansion layer uses light depth convolution to filter
Stick (NCS) are employed in this work, in order to features as source of non-linearity, in [5], suggests that this
accelerate the computing process and to ease the improves performance and provides the tuition leading to
detection process considering its portability, speed, and this design.
accuracy. The precise detection range is from 84.22 Finally, our approach allows separation of the
percent to 100 percent, with an average accuracy rate of input/output domains from the expressiveness of trans
95.89 percent. formations, which provides a convenient framework for
further analysis. We measure our performance ImageNet
Keywords— Septoria, Deep learning, Raspberry pi, classification (6) COCO object detection (1) VOC image
CNN, Intel Movidius Neural Computing Stick (NCS) segmentation.

I. INTRODUCTION III. PROPOSED APPROACH


In this task we will focus on the project called " Store The input from the system is the image captured by the
Activity Recognition Based on Raspberry Pi”. This project camera. While the output is in the form of an image that has
aims to know the number of visitors to a supermarket and been labeled or boxed whether it is identified as entering the
detect customer interest in purchasing groceries. One of the shopping area. The system can recognize how many visitors
shopping goods logistics management strategies and visit the shopping area in real time with the help of datasets.
commodity layout to increase demand from customers. The dataset used in this study is a collection of digital
images obtained from the COCO Dataset. This dataset is
In this study, the authors propose a hardware and then used to train the system to recognize human
software design that can run on the Raspberry Pi system. movements and objects. This study uses MobilenetV1 and a
Therefore, the prototype is designed using a lightweight and bounding box to recognize human movement quite well.
simple architecture using ssdmobilenetv2, a Bounding Box
Fig 1. Model optimization flow using the OpenVINO
Gambar diatas salah.

Fig.3. The proposed hardware components.

The system is run by receiving inputs (input) in the


form of a tomato leaf image acquired from the camera. The
results of the detection of septoria leaf spot disease on tomato
plants can be seen in Fig. 4.

Fig 2. Tomato normal leaf and septoria spot leaf


Disease.

This study selects a model that was trained using


Fig.4. The result
InceptionV3 [6]. There are several things proposed, such as
the concept of RMSProp Optimizer and Label Smoothing. The percentage of detection accuracy can achieve
The concept of label smoothing is an attempt to prevent between 84.22% to 100%. The calculation results show that
overfitting, in which a regularizing component will be added the average detection accuracy obtained with 30 samples of
to the loss formula. The purpose of the inception model is to tomato leaves is 95.85%.
act as a multi-level feature extractor by calculating filter
V. CONCLUSION
convolutions in the same module. The results of these filters
are then stacked into the channel dimensions before being This research has succeeded in creating a deep learning
system to detect septoria leaf spot disease on tomato plants
inserted into the next layer [6]. The model that has been
based on Intel NCS and using the InceptionV3 model. The
successfully trained will be converted into an Intermediate existing problems that affect the detection accuracy are the
Representation (IR) using OpenVINO. Ultimately, this intensity of the lighting when the image is acquired and the
model can be used on the OpenVINO inference engine distance of the camera to the object being captured.
inside Intel Movidius NCS. The results of model
optimization using OpenVINO will provide an .xml file ACKNOWLEDGMENT
containing topology, layers, connectivity, and parameters as We thank the Master of Electrical Engineering Study
well as the file containing weight and bias. Program and Erasmus+ Industry 4.0 Program, for supporting
this research.
REFERENCES

[1] Verma, Shradha, Anuradha Chug, and Amit Prakash Singh.


"Prediction models for identification and diagnosis of tomato plant
diseases." 2018 International Conference on Advances in Computing,
Communications and Informatics (ICACCI). IEEE, 2018.
[2] Brahimi, Mohammed, Kamel Boukhalfa, and Abdelouahab
Moussaoui. "Deep learning for tomato diseases: classification and
symptoms visualization." Applied Artificial Intelligence 31.4 (2017):
299-315..
[3] Fuentes, Alvaro, et al. "A robust deep-learning-based detector for
real-time tomato plant diseases and pests recognition." Sensors 17.9
(2017): 2022.
[4] Durmuş, Halil, Ece Olcay Güneş, and Mürvet Kırcı. "Disease
IV. EXPERIMENTAL RESULTS detection on the leaves of the tomato plants by using deep
The resulting system is used to detect the number of learning." 2017 6th International Conference on Agro-
Geoinformatics. IEEE, 2017.
visitors to a store based on a collection of images that have
been trained in the previous step.
[5] Rangarajan, Aravind Krishnaswamy, Raja Purushothaman, and [6] C. Szegedy, V. Vanhoucke, S. Ioffe, J. Shlens and Z. Wojna,
Aniirudh Ramesh. "Tomato crop disease classification using pre- "Rethinking the inception architecture for computer vision," in
trained deep learning algorithm." Procedia computer science 133 Proceedings of the IEEE conference on computer vision and pattern
(2018): 1040-1047. recognition, Las Vegas, Nevada, 2016.

You might also like