Using artificial neural networks to determine the prospects of using hybrid tree clones for plantation reforestation

Оценка перспективности использования клонов гибридных пород древесины является одной из актуальных задач для повышения эффективности плантационного лесовыращивания. Одним из перспективных путей решения данной задачи является применение искусственных нейронных сетей (ИНС). Настоящая научная работа является одной из немногих, где применяется ИНС для решения подобных задач в лесном хозяйстве. Для обучения нейронных сетей и определения перспективности использования клонов гибридных пород древесины для плантационного лесовыращивания были взяты биометрические данные клонов гибридной осины 2018 г. В ходе выполнения работы были построены две ИНС, где архитектура первой сети включает входной слой из 3 нейронов, 1 скрытый слой с 6 нейронами и выходной слой из 1 нейрона; архитектура второй сети включает в себя входной слой из 3 нейронов, 2 скрытых слоя по 6 нейронов и выходной слой из 1 нейрона, в которые были загружены нормализованные исходные биометрические данные для обучения определения перспективности использования клонов гибридных пород древесины для плантационного лесовыращивания. По результатам данного исследования была составлена сравнительная характеристика точности ИНС 1 и ИНС 2, которая показала, что ИНС 1 более точная, так как её отклонение на 3,49% меньше ИНС 2. Результаты настоящей работы подтвердили перспективность применения ИНС для оценки использования клонов гибридных пород древесины для плантационного лесовыращивания. По оценке расчётной перспективности ИНС 1 для плантационного лесовыращивания были выявлены клоны гибридных пород древесины VTI, ESCH3, ESCH5. Внедрение ИНС в отрасль лесного хозяйства упрощает оценку результатов биометрических показателей древесины, особенно для начинающих специалистов, что обеспечивает последующую точную оценку перспективности пород древесины. Assessing the prospects of using hybrid wood clones is one of the urgent tasks to improve the efficiency of plantation silviculture. One of the promising ways to solve this problem is the use of artificial neural networks (ANN). This research work is one of the few where ANN are used to solve such problems in forestry. Biometric data from 2018 hybrid aspen clones were taken to train neural networks and determine the potential use of hybrid wood clones for plantation silviculture. During this work, two ANNs were constructed where the architecture of the first network includes an input layer of 3 neurons, 1 hidden layer with 6 neurons and an output layer of 1 neuron, the architecture of the second network includes an input layer of 3 neurons, 2 hidden layers of 6 neurons and an output layer of 1 neuron, into which the normalized input biometric data were loaded for learning to determine the prospective use of hybrid wood species clones for plantation silviculture. Based on the results of this study, a comparison of the accuracy of ANN 1 and ANN 2 was made, which showed that ANN 1 was more accurate because its bias was 3,49% less than ANN 2. The results of this work confirmed the promise of using ANN to evaluate the use of hybrid wood clones for plantation reforestation. According to the evaluation of the calculated promisingness of ANN 1 for plantation silviculture, VTI, ESCH3 and ESCH5 hybrid wood clones were identified. The introduction of ANN in the forestry industry simplifies the evaluation of wood biometric results, especially for beginners, which provides a subsequent accurate assessment of the perspective of wood species.

A fully supervised universal adversarial perturbations and the progressive optimization

Universal Adversarial Perturbations(UAPs), which are image-agnostic adversarial perturbations, have been demonstrated to successfully deceive computer vision models. Proposed UAPs in the case of data-dependent, use the internal layers’ activation or the output layer’s decision values as supervision. In this paper, we use both of them to drive the supervised learning of UAP, termed as fully supervised UAP(FS-UAP), and design a progressive optimization strategy to solve the FS-UAP. Specifically, we define an internal layers supervised objective relying on multiple major internal layers’ activation to estimate the deviations of adversarial examples from legitimate examples. We also define an output layer supervised objective relying on the logits of output layer to evaluate attacking degrees. In addition, we use the UAP found by previous stage as the initial solution of the next stage so as to progressively optimize the UAP stage-wise. We use seven networks and ImageNet dataset to evaluate the proposed FS-UAP, and provide an in-depth analysis for the latent factors affecting the performance of universal attacks. The experimental results show that our FS-UAP (i) has powerful capability of cheating CNNs (ii) has superior transfer-ability across models and weak data-dependent (iii) is appropriate for both untarget and target attacks.

Angle of attack prediction using recurrent neural networks in flight conditions with faulty sensors in the case of F-16 fighter jet

The angle of attack (AOA) is one of the critical parameters in a fixed-wing aircraft because all aerodynamic forces are functions of the AOA. Most methods for estimation of the AOA do not provide information on the method’s performance in the presence of noise, faulty total velocity measurement, and faulty pitch rate measurement. This paper investigates data-driven modeling of the F-16 fighter jet and AOA prediction in flight conditions with faulty sensor measurements using recurrent neural networks (RNNs). The F-16 fighter jet is modeled in several architectures: simpleRNN (sRNN), long-short-term memory (LSTM), gated recurrent unit (GRU), and the combinations LSTM-GRU, sRNN-GRU, and sRNN-LSTM. The developed models are tested by their performance to predict the AOA of the F-16 fighter jet in flight conditions with faulty sensor measurements: faulty total velocity measurement, faulty pitch rate and total velocity measurement, and faulty AOA measurement. We show the model obtained using sRNN trained with the adaptive momentum estimation algorithm (Adam) produces more exact predictions during faulty total velocity measurement and faulty total velocity and pitch rate measurement but fails to perform well during faulty AOA measurement. The sRNN-GRU combinations with the GRU layer closer to the output layer performed better than all the other networks. When using this architecture, the correlation and mean squared error (MSE) between the true (real) value and the predicted value during faulty AOA measurement increased by 0.12 correlation value and the MSE decreased by 4.3 degrees if one uses only sRNN. In the sRNN-GRU combined architecture, moving the GRU closer to the output layer produced a model with better predicted values.

Application of neural network techniques to predict the heavy metals in acid mine drainage from South African mines

Acid Mine Drainage (AMD) is the formation and movement of highly acid water rich in heavy metals. Prediction of heavy metals in the AMD is important in developing any appropriate remediation strategy. This paper attempts to predict heavy metals in the AMD (Zn, Fe, Mn, Si and Ni) from South African mines using Neural Network (NN) techniques. The Backpropagation (BP) neural network model has three layers with the input layer (pH, SO42− and TDS) and output layer (Cu, Fe, Mn and Zn). After BP training, the NN techniques were able to predict heavy metals in AMD with a tangent sigmoid transfer function (tansig) at hidden layer with 5 neurons and linear transfer function (purelin) at output layer. The Levenberg-Marquardt back-propagation (trainlm) algorithm was found as the best of 10 BP algorithms with mean-squared error (MSE) value of 0.00041 and coefficient of determination (R) for all (training, validation and test) value of 0.99984. The results indicate that NN can be considered as an easy and cost-effective technique to predict heavy metals in the AMD.

Wastewater treatment by adsorption process on mineral actived carbon: modeling and prediction using an intelligent artificial approach

Currently there are several wastewater treatments processes, and several adsorbent materials consist of separating and purifying the various industrial effluents. In this work an artificial neural network (ANN) was developed to describe the dynamic adsorption of sodium decanesulfonate using actived carbon obtained by the calcination of mineral biomass under different conditions. Three inputs (time, mass of adsorbent and fixed bed height) were used in the input layer, three neurons in the hidden layer and one in the output layer for the reduced concentration. The Levenberg Marquardt back-propagation algorithm was applied. The tangent sigmoid and linear transfer functions are used for the hidden layer and the output layer respectively. The results showed a correlation coefficient R2 = 0.9965 with root mean squared error RMSE = 0.0276. An interpolation and an extrapolation stage are made to test the accuracy of the network. The results showed a high correlation coefficient R2 = 0.9969 and 0.984 respectively for the interpolation and the extrapolation. These results show the robustness and the high capacity of ANN to describe the dynamic adsorption of sodium decanesulfonate onto actived carbon.

Self-Organizing Representation Learning

Using data augmentation techniques, unsupervised representation learning methods extract features from data by training artificial neural networks to recognize that different views of an object are just different instances of the same object. We extend current unsupervised representation learning methods to networks that can self-organize data representations into two-dimensional (2D) maps. The proposed method combines ideas from Kohonen’s original self-organizing maps (SOM) and recent development in unsupervised representation learning. A ResNet backbone with an added 2D <i>Softmax</i> output layer is used to organize the data representations. A new loss function with linear complexity is proposed to enforce SOM requirements of winner-take-all (WTA) and competition between neurons while explicitly avoiding collapse into trivial solutions. We show that enforcing SOM topological neighborhood requirement can be achieved by a fixed radial convolution at the 2D output layer without having to resort to actual radial activation functions which prevented the original SOM algorithm from being extended to nowadays neural network architectures. We demonstrate that when combined with data augmentation techniques, self-organization is a simple emergent property of the 2D output layer because of neighborhood recruitment combined with WTA competition between neurons. The proposed methodology is demonstrated on SVHN and CIFAR10 data sets. The proposed algorithm is the first end-to-end unsupervised learning method that combines data self-organization and visualization as integral parts of unsupervised representation learning.

PATTERN RECOGNITION AKSARA LAMPUNG MENGGUNAKAN ALGORITMA NEURAL NETWORK

Indonesia merupakan Negara yang terdiri dari berbagai macam suku dan budaya, Indonesia juga memiliki berbagai macam bahasa daerah, salah satunya merupakan bahasa Lampung. Bahasa Lampung merupakan bahasa asli suku lampung, didalam bahasa lampung terdapat aksara yaitu aksara lampung. Aksara lampung memiliki 20 kepala bahasa dan 12 tanda baca. Pada penelitian ini dilakukan analisa pengenalan tulisan berdasarkan perubahan iterasi dengan menggunakan metode neural network. Neural network merupakan jaringan saraf yang terdiri dari unit dasar yang seperti analog dengan neuron, neural network dibagi berdasarkan 3 layer yaitu input layer, hidden layer dan output layer. Dimana setiap node pada masing-masing layer memiliki suatu error rate, yang akan digunakan untuk proses training. Pada penelitian ini akan menggunakan bahasa pemograman python. Percobaan untuk induk surat akan menggunakan 20 huruf aksara lampung dengan masing-masing huruf terdapat 10 pengujian citra, dan percobaan untuk anak surat akan menggunakan 12 huruf anak aksara lampung dengan masing-masing huruf terdapat 10 pengujian citra, sehingga total keseluruhan dataset mejadi 320 citra. Hasil yang diperoleh dari proses pemeriksaan masing-masing adalah 75%, untuk induk surat dengan sebaran 135 citra terdeteksi benar dan 45 citra tidak terdeteksi dengan benar. Untuk anak surat 81 citra terdeteksi dengan benar dan 27 citra tidak terdeteksi dengan benar.

Self-Organizing Representation Learning

Using data augmentation techniques, unsupervised representation learning methods extract features from data by training artificial neural networks to recognize that different views of an object are just different instances of the same object. We extend current unsupervised representation learning methods to networks that can self-organize data representations into two-dimensional (2D) maps. The proposed method combines ideas from Kohonen’s original self-organizing maps (SOM) and recent development in unsupervised representation learning. A ResNet backbone with an added 2D <i>Softmax</i> output layer is used to organize the data representations. A new loss function with linear complexity is proposed to enforce SOM requirements of winner-take-all (WTA) and competition between neurons while explicitly avoiding collapse into trivial solutions. We show that enforcing SOM topological neighborhood requirement can be achieved by a fixed radial convolution at the 2D output layer without having to resort to actual radial activation functions which prevented the original SOM algorithm from being extended to nowadays neural network architectures. We demonstrate that when combined with data augmentation techniques, self-organization is a simple emergent property of the 2D output layer because of neighborhood recruitment combined with WTA competition between neurons. The proposed methodology is demonstrated on SVHN and CIFAR10 data sets. The proposed algorithm is the first end-to-end unsupervised learning method that combines data self-organization and visualization as integral parts of unsupervised representation learning.

Identification of Microcontroller Based Objects Using Image Classification in Python

In this paper, I built an AI model using deep learning, which identifies whether a given image is of an Arduino, a Beaglebone Black or a Jetson Nano. The identification of the object is based on prediction. The model is trained using 300 to 350 datasets of each category and is tested multiple times using different images at different angles, background colour and size. After multiple testing, the model is found to have 95 percent accuracy. Model used is Sequential and uses Convolution Neural Network (CNN) as its architecture. The activation function of each layer is RELU and for the output layer is Softmax. The output is a prediction and hence it is of probability type. This is a type of an application based project. The entire scripting is done using Python 3 programming language. Keywords: image classification, microcontroller boards, python, AI, deep learning, neural network

Prediksi dan Klasifikasi Buku Menggunakan Metode Backpropagation

Students of SMP Negeri 2 Lengayang have different interests in determining the books they are interested in, so that the library often has difficulty determining the books that are most entered by students, this is because they have not used the right system in determining the type and number of books, only based on the estimated number. Students and subjects only, as a result school students stock books of the books they want to borrow. Based on the above, a method is needed to predict and classify the amount of book stock in the future. The data used is a recap of monthly book lending, from 2018 to 2020 in the third month, with a total of 1653 transactions and 5 types of books processed, then the data is analyzed using the Backpropogation method. The results obtained are using a 5-3-1 pattern with a learning rate of 0.01, a goal of 0.01, the number of input units for the Weapon layer 5, the number of units in the hidden layer and the number of output layer units that are placed on 1 layer, and to carry out training using two phases namely feedforward and backpropagation phases. It is removed from this research that the backpropagation method can provide a classification prediction of the number of books that must be provided in the following year based on the number of data entered or the number of data entered.