Is Pretraining Necessary for hyperspectral image classification?

Lee, Hyungtae; Eum, Sungmin; Kwon, Heesung

doi:10.1109/igarss.2019.8898734

Cited by 12 publications

(5 citation statements)

References 8 publications

Supporting

Mentioning

Contrasting

Unclassified

Order By: Relevance

“…Transfer learning adalah teknik di pembelajaran mesin di mana model yang telah dilatih pada tugas tertentu. Pada transfer learning beberapa layer dari model asli yang telah dilatih diubah dan digunakan sebagai pengenalan fitur yang telah dipelajari, dan lapisan pengenalan fitur tersebut dapat dihubungkan dengan beberapa layer terakhir yang baru ditambahkan dan dilatih ulang untuk tugas yang berbeda [12]. Transfer learning membantu dalam mempercepat proses pelatihan model baru dan meningkatkan akurasi model baru [13].…”

Section: Pendahuluanunclassified

Chicken Disease Detection Based on Fases Image Using EfficientNetV2L Model

Mustopa,

Sasongko,

Nawawi

et al. 2023

SISTEMASI

View full text Add to dashboard Cite

Peternakan membutuhkan inovasi teknologi untuk meningkatkan produktivitas dan efisiensi. Ayam adalah hewan ternak yang memiliki prospek pasar yang baik. Namun, tidak semua peternak memahami tentang penyakit dan tanda-tanda ayam yang sakit. Pendeteksian penyakit pada ayam dapat dilakukan dengan berbagai metode, salah satunya adalah dengan melihat dari bentuk tinja yang dihasilkan ayam. Citra pada feses dapat dideteksi dengan menggunakan machine learning. Convolutional Neural Network (CNN) digunakan untuk mempercepat prediksi penyakit. Transfer learning digunakan untuk memanfaatkan pengetahuan yang sudah dipelajari oleh model sebelumnya. Pada penelitian ini, kami mengusulkan model arsitektur CNN sendiri dan memaparkan penelitian dengan membangun sebuah model baru untuk mendeteksi dan mengklasifikasikan penyakit pada ayam melalui fesesnya. Proses pelatihan model dilakukan dengan memasukkan data latih dan data validasi, jumlah epoch, dan objek checkpointer yang sudah dibuat. Tahap mengubah hyperparameter dilakukan untuk meningkatkan tingkat akurasi model. Penelitian dilakukan dengan menguji dataset yang didapatkan dari repositori kaggle yang memiliki citra coccidiosis, salmonella, newcastle, dan feses yang sehat. Pada hasil penelitian menunjukkan bahwa model usulan hanya mendapatkan akurasi sebesar 93% sedangkan untuk akurasi terbaik pada penelitian menggunakan model EfficientNerV2L dengan optimizer RMSProp yaitu 97%.

show abstract

Section: Pendahuluanunclassified

Chicken Disease Detection Based on Fases Image Using EfficientNetV2L Model

Mustopa,

Sasongko,

Nawawi

et al. 2023

SISTEMASI

View full text Add to dashboard Cite

show abstract

“…[25] introduce a system that automatically designs a structure with the existing layers to provide the highest classification accuracy. 2 Although we elaborated such practical research questions within our preliminary work [2], we managed to analyze our pretrain-finetune approach with a limited number of experimental evaluations and thus could not address the resolutions for the questions in a thorough manner. In this journal manuscript, the entire set of experiments has been redesigned to reach at the conclusions through more comprehensive analyses.…”

Section: B Hyperspectral Image Classificationmentioning

confidence: 99%

Exploring Cross-Domain Pretrained Model for Hyperspectral Image Classification

Lee,

Eum,

Kwon

2022

Preprint

Self Cite

View full text Add to dashboard Cite

A pretrain-finetune strategy is widely used to reduce the overfitting that can occur when data is insufficient for CNN training. First few layers of a CNN pretrained on a large-scale RGB dataset are capable of acquiring general image characteristics which are remarkably effective in tasks targeted for different RGB datasets. However, when it comes down to hyperspectral domain where each domain has its unique spectral properties, the pretrain-finetune strategy no longer can be deployed in a conventional way while presenting three major issues: 1) inconsistent spectral characteristics among the domains (e.g., frequency range), 2) inconsistent number of data channels among the domains, and 3) absence of large-scale hyperspectral dataset.We seek to train a universal cross-domain model which can later be deployed for various spectral domains. To achieve, we physically furnish multiple inlets to the model while having a universal portion which is designed to handle the inconsistent spectral characteristics among different domains. Note that only the universal portion is used in the finetune process. This approach naturally enables the learning of our model on multiple domains simultaneously which acts as an effective workaround for the issue of the absence of large-scale dataset.We have carried out a study to extensively compare models that were trained using cross-domain approach with ones trained from scratch. Our approach was found to be superior both in accuracy and in training efficiency. In addition, we have verified that our approach effectively reduces the overfitting issue, enabling us to deepen the model up to 13 layers (from 9) without compromising the accuracy.

show abstract

“…For the second step, we redesigned the online hard example mining (OHEM) [4] To evaluate the proposed HEG approach, we implemented a 9-layer fully convolutional network (FCN) inspired by [5]. The FCN architecture has proved to be suitable for pixelwise classification [5]- [8]. We validate our approach to red tide detection using the large-scale remote sensing image dataset obtained from multi-spectral GOCI (Geostationary Ocean Color Imager) [9] on a geostationary satellite.…”

Section: Hard Example Candidatesmentioning

confidence: 99%

Generating Hard Examples for Pixel-Wise Classification

Lee¹,

Kwon²,

Kim³

2021

IEEE J. Sel. Top. Appl. Earth Observations Remote Sensing

Self Cite

View full text Add to dashboard Cite

Pixel-wise classification in remote sensing identifies entities in large-scale satellite-based images at the pixel level. Few fully annotated large-scale datasets for pixel-wise classification exist due to the challenges of annotating individual pixels. Training data scarcity inevitably ensues from the annotation challenge, leading to overfitting classifiers and degraded classification performance. The lack of annotated pixels also necessarily results in few hard examples of various entities critical for generating a robust classification hyperplane. To overcome the problem of the data scarcity and lack of hard examples in training, we introduce a two-step hard example generation (HEG) approach that first generates hard example candidates and then mines actual hard examples. In the first step, a generator that creates hard example candidates is learned via the adversarial learning framework by fooling a discriminator and a pixel-wise classification model at the same time. In the second step, mining is performed to build a fixed number of hard examples from a large pool of real and artificially generated examples. To evaluate the effectiveness of the proposed HEG approach, we design a 9-layer fully convolutional network suitable for pixel-wise classification. Experiments show that using generated hard examples from the proposed HEG approach improves the pixel-wise classification model's accuracy on red tide detection and hyperspectral image classification tasks.

show abstract

Is Pretraining Necessary for hyperspectral image classification?

Cited by 12 publications

References 8 publications

Chicken Disease Detection Based on Fases Image Using EfficientNetV2L Model

Chicken Disease Detection Based on Fases Image Using EfficientNetV2L Model

Exploring Cross-Domain Pretrained Model for Hyperspectral Image Classification

Generating Hard Examples for Pixel-Wise Classification

Contact Info

Product

Resources

About