









































Pa
ge

 
1



Pa
ge

 
57

American Journal of  Smart 
Technology and Solutions (AJSTS)

Evaluating Soybean Root Health Using Residual Neural Network (ReNN) 
Based Image Analysis 

Vivek Gupta1*, Jhankar Moolchadani2, Harsh Singh Chouhan2

Volume 4 Issue 2, Year 2025
ISSN: 2837-0295 (Online)

DOI: https://doi.org/10.54536/ajsts.v4i2.5382
https://journals.e-palli.com/home/index.php/ajsts

Article Information ABSTRACT

Received: June 22, 2025

Accepted: July 31, 2025

Published: September 05, 2025

ReNN’s layer-wise image segmentation and robust data processing address the complexities 
of  soybean root analysis, providing valuable insights for improved crop management. 
Accurate assessment of  soybean root health is crucial for optimal crop production and 
growth. This study utilizes Residual Neural Network (ReNN) image evaluation to analyze 
soybean root development. Field data is collected and integrated by deploying sensor-
based devices (IoT Devices) to evaluate soybean crop stages. ReNN facilitates data 
preprocessing, layer-wise image segmentation, and effective data processing, enabling 
accurate assessment of  root health, plant vigor, flower fragmentation, and fruit formation. 
This approach predicts soybean crop yield and provides valuable insights into degradation 
detection and decision-making for optimal soybean cultivation practices. ReNN utilizes 
layer-based image formation, collecting data from farm fields through sensor-based 
devices. The dataset encompasses various environmental and weather conditions, ensuring 
comprehensive coverage. Key considerations for data preprocessing include temperature, 
humidity, precipitation as the Weather conditions, soil type, moisture, sunlight as 
Environmental factors, field location, soil heterogeneity as Spatial variability, and growth 
stage, seasonality as Temporal variability. By integrating these factors, ReNN enables 
accurate evaluation of  soybean root conditions, facilitating root health assessment, Plant 
growth monitoring, Yield prediction, and optimized cultivation practices.

Keywords
Convolution Neural Network 
(CoNN), Data Preprocessing 
System (DPS), Internet of  Things 
(IoT), Neural Network (NN), 
Residual Neural Networks 
(ResNN)

1 Department of  Computer Science and Engineering, ASET & Amity University Gwalior, MP, India
2 Department of  Electronics and Communication & Indore Institute of  Science and Technology, Indore, MP, India
* Corresponding author’s e-mail: vivek.gupta5@s.amity.edu

INTRODUCTION
In the realm of  technological approaches, the domain 
of  agriculture is vast and plays a significant role in the 
advancement of  agricultural technology. The approaches 
used were used to evaluate the data set of  images. The 
images are clear with proper originality and appropriate 
for evaluation. Therefore, the technological enhancement 
toward the uses of  artificial intelligence and its based 
method for the image evaluations for the soybean crop 
farming. Technologically, the devices deployed for the 
image extractions and the images of  the plant picked 
from the field and images are further forwarded for 
the input values. Image recognition in agriculture has 
promoted research for the increase in the production 

of  the crop. Additionally, crop evaluation research 
facilitates the identification of  the health condition of  
the crop plant. The system that the paper shows focus 
on the design of  the device and the adoption of  the best 
method based on artificial intelligence for the validation 
of  image-based data sets. The approaches of  Residual 
Neural Network (RNN) are suitable for proper validation 
of  the image-based dataset. Here, the research engaged 
the IoT-based image cameras that study and evaluate data 
sets of  files and input the values of  the RNN method of  
evaluation of  the crop conditions. As research supports, 
the ideology is extracted based on technical devices that 
support the states of  the root formation and condition of  
the soybean plant in the farming of  soybeans (He, 2016).

Figure 1: Model of  the Evaluating the Soybean Plants and Roots Images Analysis

LITERATURE REVIEWS
The Literature review considers the two states of  image 
segmentation, the first is considered which is based on 

computer vision and the states of  images. The second is 
based on the model of  IoT controllers that illustrates the 
image processing for the segmentation and is forwarded 



Pa
ge

 
58

https://journals.e-palli.com/home/index.php/ajsts

Am. J. Smart. Technol. Solutions 4(2) 57-62, 2025

to the machine learning-based model of  the imaging 
system is ResNN (Krizhevsky, 2017).
Based on Computer Vision Techniques:
Computer vision technology can be used in agriculture 
to help farmers monitor crops, detect pests and diseases, 
and improve yields:

Crop Monitoring
High-resolution cameras and algorithms can analyze 
images of  crops to provide insights into plant health, 
growth stages, and potential yield.

Pest and Disease Detection
Computer vision systems can help identify potential pests 
and diseases that may cause crop losses.

Weed Identification
Computer vision can help identify weeds and apply 
precision herbicides.

Yield Prediction
By analysing factors like plant height, leaf  area, and fruit 
count, computer vision can help predict crop yield.

Automated Harvesting
Computer vision can help automate the process of  
harvesting crops.

Soil Analysis
Computer vision can help analyze soil conditions.

Nutrient Management
Computer vision can help identify nutrient deficiencies 
and apply targeted fertilizers.
Computer vision can be used in a variety of  ways, 
including: (Leibe et al., 2016)

1. UAVs: Unmanned aerial vehicles (UAVs) equipped 
with computer vision systems can help farmers monitor 
crops and assess plant health.

2. Mobile robots: Farmers can use mobile robots 
equipped with computer vision to drive around and 
collect data.

3. Static cameras: Farmers can use static cameras to 
take images of  crops from an advantageous position. 

The Authors Suggested that, Based on Data Availability
Image Quality
Many studies and competitions used plant image datasets 
with a single background. For example, the Plant Village 
dataset contains many labelled plants leaf  images from 
various species with different diseases, but the pictures were 
from a controlled environment, and their backgrounds 
are very simple. However, because of  lighting, occlusion, 
and shadows in the natural atmosphere, the image quality 
and visual perception ability will degenerate greatly. Many 
noises appear in the images, which is a big challenge for 
automatically analysing unconstrained natural images in 
the field. Although human visual systems can easily deal 

with these problems, establishing a computational model 
of  plant phenotyping is still an open-ended question (Hu 
et al., 2018). 

Image Annotation
Deep learning needs to learn features from sufficient 
annotated data, but data annotation faces the following 
challenges:

a. Manual annotation sometimes requires a large 
amount of  prior or professional domain knowledge and 
rich working experience.

b. Data annotation is a time-consuming and hectic step, 
especially in object detection and image segmentation. 
Detection and segmentation require instance-level (boxes) 
and pixel-level annotations (masks). If  more and more 
images are to be annotated, the workload will be massive, 
while efficiency and accuracy cannot be guaranteed (Lin  
et al., 2019).

c. Some images lack visual cues, such as hyperspectral 
and thermal imaging, so it is much more difficult to label 
these data than RGB images. 
Online researchers have deployed a human-machine 
collaboration interface called fluid annotation (Andriluka 
et al., 2018) that can be used to annotate the class label and 
delineate the contours of  every object and background in 
an image (Bello et al., 2019). 

The Authors Suggested the Data-Based Analysis
Algorithm Robustness
At present, some mainstream algorithms perform well on 
particular datasets, and most of  them are only designed for 
specific organs or specific plant species. Due to the large 
differences in colour, shape, size, and other characteristics 
between different detection objects, these algorithms do 
not generalize well. When the dataset changes, many 
algorithms will be invalid, so researchers must redesign 
the feature extractor and readjust the hyperparameters. 
For stress phenotyping, the degree of  plant stress changes 
over time. The model needs to be improved and modified 
to be dynamically analysed throughout the entire cycle of  
stress, which is a challenge for designing a processing 
framework (Li et al., 2018). 

Deep Learning
Firstly, deep learning-based algorithms rely on a big 
number of  labelled sample images, which makes it difficult 
to achieve excellent results in the following three scenarios:

• Training samples do not exist in some object categories.
• There are a few samples in object categories.
• The sample size of  different categories is extremely 

imbalanced. Then, some deep learning-based solutions 
lack prior knowledge, and it is difficult to adaptively use 
my discriminative visual features. Moreover, the deep 
neural network is used as a “black box”, which cannot 
perform explicit reasoning and lacks interpretability. Tasks 
like gene-phenotype association and image description 
require high-level logical reasoning and often can’t be 
solved with simple classification or regression methods. 



Pa
ge

 
59

https://journals.e-palli.com/home/index.php/ajsts

Am. J. Smart. Technol. Solutions 4(2) 57-62, 2025

These problems need more advanced approaches, such 
as: Deep learning model, Multimodal learning, Graph-
based methods (Zagoruyko et al., 2017).
Finally, most 3D point clouds are still analysed by utilizing 
traditional 3D processing methods. Solutions based 
on deep learning have not been popularized in plant 
phenotyping. The following research aspects are worthy 
of  attention in the future.

1. Plant images with complex backgrounds require 
effective segmentation of  the foreground and 
background. Methods based on deep learning are very 
suitable for image segmentation, but image annotation 
becomes the major limiting factor for applying 
deep learning in plant phenotyping. To reduce the 
requirements of  annotated data, the following solutions 
were proposed and developed: On the one hand, some 
image generation strategies (e.g., GANs) can be applied 
to increase image diversity and availability. On the other 
hand, the dependence of  models on data can be reduced 
by improving algorithms, such as zero sample learning, 
small sample learning, transfer learning, and so on (Tan 
& Le, 2019). 

2. Most existing deep learning algorithms rely on many 
labelled images to fit many parameters for prediction, 
ignoring the prior knowledge of  many domain 
associations and the intuitive understanding of  decision-
making processes, which limits the interpretation of  
model functions to a certain extent (Pharm, 2020). 

3. CNN has great potential in 3D reconstruction and 
segmentation. Some approaches use CNN to project 2D 
segments onto 3D representations or apply them to 3D 
images directly. Thus, a lot of  3D processing work requires 
to application of  CNN architectures to characterize and 
understand plant phenotypes directly

Model of  IOT devices for Image Extraction
The author suggested that the rapid evolution of  IoT 
devices necessitates efficient image super-resolution 
techniques, while existing advanced methods, based on 
deep convolutional neural networks, are too resource-
intensive for these circuit-based models, and this gap 
illustrates the need for a more suitable solution. In this 
study, we introduce a lightweight, essentially super-
resolution model specially designed for IoT devices. 
This model incorporates a novel deep residual feature 
distillation block (DRFDB), which leverages a depth-
wise-separable convolution block (DCB) for effective 
feature extraction. Determined to reduce computational 
and memory demands without changes to image quality. 
The model shows improved performance metrics like 
PSNR, while requiring fewer parameters and less memory 
usage, making it highly suitable for IoT applications. This 
study presents a breakthrough in super-resolution for IoT 
devices, balancing high-quality image reconstruction with 
the limited resources of  these devices (Gao et al., 2019).

MATERIALS AND METHODS
ResNN is an artificial Intelligence method to help in the 

evaluation of  the images of  the plant soybean crop. A 
Residual Neural Network (ResNet) stacks residual blocks 
on top of  each other to form a network.

The Residual Neural Network 
To know about residual neural networks and the most 
popular ResNets, including ResNet-34, ResNet-50, 
and ResNet-101. In current years, the field of  artificial 
intelligence applied to computer vision has undergone 
far-reaching transformations due to the introduction of  
new technologies (Xie S, Zerhouni E, Huang G, 2017). 

a. The rapid progress in deep learning has enabled 
computer vision models to achieve unprecedented 
levels of  accuracy and efficiency in tasks such as image 
recognition, object detection, and face recognition, 
surpassing human capabilities in many cases.

b. But, while it gives us the option of  adding more fully 
connected layers to the CNNs to solve more complicated 
tasks in computer vision, it comes with its own set of  
issues. It has been observed that training the NN becomes 
more difficult with the extension in the number of  added 
layers, and in some cases, the accuracy dwindles as well.

c. It is here that the use of  ResNet assumes importance. 
Deeper neural networks are tough to train. With ResNet, 
it becomes easy to surpass the difficulties of  training very 
deep neural networks.

d. When working with deep convolutional neural 
networks to break a problem related to computer vision, 
machine learning experts engage in mounding further 
layers. These fresh layers help break down complex 
problems more efficiently, as the different layers can be 
trained for varying tasks to get largely accurate results. 

e. While the number of  piled layers can enrich the 
features of  the model, a deeper network can show the 
issue of  declination. Basically, as the number of  layers of  
the neural network increases, the complexity of  situations 
may get impregnated and sluggishly degrade after a point. 
As a result, the performance of  the model deteriorates 
both on the training and testing data.

f. This declination isn’t a result of  overfitting. Rather, it 
may affect the initialization of  the network, optimization 
function, or, more importantly, the problem of  
evaporating or exploding slant

Various Factors Based on Types of  Resnet Are as 
Follows
ResNet-50 Architecture: Bottleneck Design
ResNet-50 uses a bottleneck design, which reduces the 
number of  parameters and computational cost. 3-layer 
blocks: ResNet-50 uses a stack of  3 layers instead of  the 
earlier 2-layer blocks, forming a 3-layer bottleneck block. 
Higher accuracy: ResNet-50 achieves much higher accuracy 
than the 34-layer ResNet model. Performance: The 50-layer 
ResNet-50 achieves a performance of  3.8 billion FLOPS.

ResNet-101 and ResNet-152 Architecture: Large 
Residual Networks
ResNet-101 and ResNet-152 are constructed using more 



Pa
ge

 
60

https://journals.e-palli.com/home/index.php/ajsts

Am. J. Smart. Technol. Solutions 4(2) 57-62, 2025

than 3-layer blocks, enabling deeper networks with lower 
complexity. Lower complexity: Despite increased network 
depth, the 152-layer ResNet has much lower complexity 
(11.3 billion FLOPS) than VGG-16 or VGG-19 nets 
(15.3/19.6 billion FLOPS).

ResNet-50 with Keras: Keras API
Keras is a popular deep learning API known for its 
simplicity and ease of  use. Pre-trained models: Keras 
comes with several pre-trained models, including 
ResNet-50, which can be used for various experiments 
and applications.

Technical Approaches for Validation of  the Soybean 
Plant and the Condition of  Its Roots
The technology of  computer vision and the integration 
of  the model of  ResNet is being applied to the farming 

of  soybean crops. The adjustment design for the 
evaluation supports and recognizes the image data 
sets and validates the progress of  the soybean plants. 
Recommended the IoT and ResNet model based on 
a computer vision system is as below block diagram, 
where the camera is built for extracting the images, 
and after those images are forwarded to the memory 
shuttle of  memory, which is built into the circuit of  
IoT devices. The memory transferring the images to 
the computer system, where an algorithm extracts the 
image-based data of  the sets that are recognized for 
the image segmentation. The Block Diagram shows 
the stepwise uses of  the IOT model and the process 
of  input values. The block diagram also integrates the 
steps of  preprocessing with the help of  modules of  
ResNN (Simonyan & Zisserman, 2014; Szegedy et al., 
2015).

Figure 2: Model-Based to define the IOT and ResNN for Validating Data Set

RESULT AND DISCUSSION
Evaluation And Analysis Of  Soybean Roots Using 
Resnn
Technology ReNN Architecture Consists of
Residual Block
Residual blocks are the main components of  the Residual 
Neural Network. In a classical neural network, the input 
is transformed by a set of  convolutional layers then it is 
passed to the activation function. In a residual network, 
the input to the block is added to the output of  the block, 
creating a residual connection. The output of  the residual 
block H(xi) can be represented by:
H(xi) = F(xi) + xi
F(xi) represents the residual mapping learned by the 
network. The presence of  the identity term x allows the 
gradient to flow more easily.

S Connection
S connection is a Skip connection that helps in forming the 
residual blocks. Skip connection consists of  the input of  
the residual block that is bypassed over the convolutional 
layer and added to the output of  the residual block.

ST Layers
ResNet architectures are formed by stacking multiple 
residual blocks together. Using these multiple residual 
blocks together, resnet architecture can be built very 
deep. Versions of  ResNet with 50,101,152 layers were 
introduced.

Global Average Pooling(GAP) 
Resnet architectures typically utilize Global Average 
Pooling as the final layer before the fully connected layer.
GAP reduces spatial dimensions to a single value per 
feature map, providing a compact representation of  the 
entire feature map (Zhang et al., 2019).
Healthy soybean roots have the following characteristics 
(Chen et al., 2020)

• Depth: Soybean roots typically grow to a depth of  
2–3 feet, but most of  the roots are in the top 6–12 inches 
of  soil.

• Nodules: A healthy soybean plant should have 6–20 
large nodules on the main tap root and smaller nodules 
on the auxiliary roots. Nodules are formed by bacteria in 
the soil and provide much of  the plant’s nitrogen supply.



Pa
ge

 
61

https://journals.e-palli.com/home/index.php/ajsts

Am. J. Smart. Technol. Solutions 4(2) 57-62, 2025

• Colour: A healthy nodule is red or pink, which 
indicates active nitrogen fixation. A white or gray nodule 
is immature and should be checked again in a week. A 
green, brown, or mushy nodule is dead.

• Root type: Soybeans have both deep, vertical roots 
and shallow, lateral roots. The deep roots access water 
from deeper soil layers, while the shallow roots increase 
the plant’s ability to absorb nutrients from the topsoil.

Figure 3: Images of  Standardized Formation of  Soybean Plant 

Y (output) = x+F(x)

ReNN Model for Evaluation and Image-Based 
Analysis

the effective and productive extraction of  the root image, 
so the progress and evaluation of  the roots help and are 
more supportive of  the progress of  the plant. That plant’s 
progress directly raises the production of  the soybean.
Here are illustrated the key facts for ResNN, which is 
supportive of  the evaluation of  the soybean plants in the 
future study and project of  IoT deployments.

• ResNets (Residual Networks) are a variant of  deep 
learning algorithms that are particularly for image 
recognition and processing tasks. ResNets are known for 
their make to train very deep networks without overfitting

• ResNets are helpful for detection tasks. Key point 
detection is the task of  locating points on an object in an 
image. For example, detection can be used to locate the 
eyes, nose, and mouth on a human face.

• ResNets are well-suited because they can learn to 
extract from images at different scales.

REFERENCES
Bello, I., Zoph, B., Vaswani, A., Shlens, J., & Le, Q. V. 

(2019). Attention augmented convolutional networks. 
In Proceedings of  the IEEE/CVF International Conference 
on Computer Vision (pp. 3286–3295). IEEE.

Chen, B., Ghiasi, G., Liu, H., Lin, T.-Y., Kalenichenko, D., 
Adam, H., & Le, Q. V. (2020). MnasFPN: Learning 
latency-aware pyramid architecture for object 
detection on mobile devices. In Proceedings of  the 
IEEE/CVF Conference on Computer Vision and Pattern 
Recognition (pp. 13607–13616). IEEE.

Deng, J., Dong, W., Socher, R., Li, L.-J., Li, K., & Fei-Fei, 
L. (2009, June 20). ImageNet: A large-scale hierarchical 
image database. In 2009 IEEE Conference on Computer 
Vision and Pattern Recognition (pp. 248–255). IEEE.

DeVries, T., & Taylor, G. W. (2017, August 15). Improved 
regularization of  convolutional neural networks with 

Considering, 
Farming Side s1: Images Inputs Ms : Length T1 : Root 
size R_Lat, R_Nod, R_Bra
In the process of  ResNN, the extraction of  images is 
consolidated from the M1 position to the Ms position. 
After that, the land-side-wise extraction is in observation 
for the same, considering the s1 to s nth side. Whereas the 
root observation key points are evaluated and analyzed 
to add f(x) = T1+R_Lat, similarly add f(y)= T1+R_Nod 
and f(z)= T1+R_Bra. 
The functions f(x) = partially integrated mapping and 
putting the evaluation steps for States Lateral Roots, 
States Nodules, and States Branchtop root.
Therefore, Output Evaluation y = f( x ) +f( y )+f( z )
Same way, the changes in image observation, the values 
are integrated, and the analysis of  the present image data 
set would be responsible for the analysis of  the healthy 
condition of  the roots, and also changes in the image data 
set to observe the desired diseases and progress of  the plant.
 
CONCLUSION
The Observation implementation is still in process for 



Pa
ge

 
62

https://journals.e-palli.com/home/index.php/ajsts

Am. J. Smart. Technol. Solutions 4(2) 57-62, 2025

cutout. arXiv preprint arXiv:1708.04552.
Farooq, M., & Hafeez, A. (2020, March 31). Covid-resnet: 

A deep learning framework for screening of  COVID-19 from 
radiographs. arXiv preprint arXiv:2003.14395.

Gao, S. H., Cheng, M. M., Zhao, K., Zhang, X. Y., Yang, 
M. H., & Torr, P. H. S. (2021). Res2Net: A new multi-
scale backbone architecture. IEEE Transactions on 
Pattern Analysis and Machine Intelligence, 43(2), 652–662.

He, K., Zhang, X., Ren, S., & Sun, J. (2016). Deep 
residual learning for image recognition. In Proceedings 
of  the IEEE Conference on Computer Vision and Pattern 
Recognition (pp. 770–778). IEEE.

Hu, J., Shen, L., & Sun, G. (2018). Squeeze-and-excitation 
networks. In Proceedings of  the IEEE Conference on Computer 
Vision and Pattern Recognition (pp. 7132–7141). IEEE.

Huang, G., Liu, Z., Van Der Maaten, L., & Weinberger, K. 
Q. (2017). Densely connected convolutional networks. 
In Proceedings of  the IEEE Conference on Computer Vision 
and Pattern Recognition (pp. 4700–4708). IEEE.

Krizhevsky, A., Sutskever, I., & Hinton, G. E. (2017). 
ImageNet classification with deep convolutional neural 
networks. Communications of  the ACM, 60(6), 84–90.

Leibe, B., Matas, J., Sebe, N., & Welling, M. (Eds.). (2016, 
September 16). Computer Vision–ECCV 2016: 14th 
European Conference, Amsterdam, The Netherlands, October 
11–14, 2016, Proceedings, Part IV. Springer.

Li, Z., Sang, N., Chen, K., Gao, C., & Wang, R. (2018, 
March 8). Learning deep features with adaptive triplet 
loss for person reidentification. In MIPPR 2017: 
Pattern Recognition and Computer Vision (Vol. 10609, pp. 
90–95). SPIE.

Lin, S., Ji, R., Yan, C., Zhang, B., Cao, L., Ye, Q., Huang, F., 
& Doermann, D. (2019). Towards optimal structured 
CNN pruning via generative adversarial learning. In 
Proceedings of  the IEEE/CVF Conference on Computer 
Vision and Pattern Recognition (pp. 2790–2799). IEEE.

Nam, H., Ha, J. W., & Kim, J. (2017). Dual attention 
networks for multimodal reasoning and matching. In 
Proceedings of  the IEEE Conference on Computer Vision 
and Pattern Recognition (pp. 299–307). IEEE.

Pham, T. D. (2020). A comprehensive study on 
the classification of  COVID-19 on computed 
tomography with pre-trained convolutional neural 
networks. Scientific Reports, 10(1), 1–8.

Roy, S. K., Manna, S., Song, T., & Bruzzone, L. (2020). 
Attention-based adaptive spectral-spatial kernel ResNet 
for hyperspectral image classification. IEEE Transactions 

on Geoscience and Remote Sensing, 59(9), 7831–7843.
Simonyan, K., & Zisserman, A. (2014, September 4). 

Very deep convolutional networks for large-scale 
image recognition. arXiv preprint arXiv:1409.1556.

Szegedy, C., Liu, W., Jia, Y., Sermanet, P., Reed, S., Anguelov, 
D., Erhan, D., Vanhoucke, V., & Rabinovich, A. 
(2015). Going deeper with convolutions. In Proceedings 
of  the IEEE Conference on Computer Vision and Pattern 
Recognition (pp. 1–9). IEEE.

Tan, M., & Le, Q. (2019, May 24). EfficientNet: 
Rethinking model scaling for convolutional neural 
networks. In International Conference on Machine Learning 
(pp. 6105–6114). PMLR.

Tseng, H. Y., Lee, H. Y., Huang, J. B., & Yang, M. H. 
(2020). Cross-domain few-shot classification via 
learned feature-wise transformation. arXiv preprint 
arXiv:2001.08735.

Wang, Z., Yu, Z., Zhao, C., Zhu, X., Qin, Y., Zhou, Q., 
Zhou, F., & Lei, Z. (2020). Deep spatial gradient and 
temporal depth learning for face anti-spoofing. In 
Proceedings of  the IEEE/CVF Conference on Computer 
Vision and Pattern Recognition (pp. 5042–5051). IEEE.

Xie, S., Girshick, R., Dollár, P., Tu, Z., & He, K. (2017). 
Aggregated residual transformations for deep neural 
networks. In Proceedings of  the IEEE Conference on Computer 
Vision and Pattern Recognition (pp. 1492–1500). IEEE.

Zagoruyko, S., & Komodakis, N. (2016, May 23). Wide 
residual networks. arXiv preprint arXiv:1605.07146.

Zerhouni, E., Lányi, D., Viana, M., & Gabrani, M. 
(2017, April 18). Wide residual networks for mitosis 
detection. In 2017 IEEE 14th International Symposium 
on Biomedical Imaging (ISBI) (pp. 924–928). IEEE.

Zhang, H., Wu, C., Zhang, Z., Zhu, Y., Lin, H., Zhang, 
Z., Sun, Y., He, T., Mueller, J., Manmatha, R., & Li, 
M. (2022). ResNeSt: Split-attention networks. In 
Proceedings of  the IEEE/CVF Conference on Computer 
Vision and Pattern Recognition (pp. 2736–2746). IEEE.

Zhang, K., Sun, M., Han, T. X., Yuan, X., Guo, L., & Liu, 
T. (2017). Residual networks of  residual networks: 
Multilevel residual networks. IEEE Transactions on 
Circuits and Systems for Video Technology, 28(6), 1303–1314.

Zhong, X., Gong, O., Huang, W., Li, L., & Xia, H. (2019, 
September 22). Squeeze-and-excitation wide residual 
networks in image classification. In 2019 IEEE 
International Conference on Image Processing (ICIP) (pp. 
395–399). IEEE.


