Abdelilah Ganmati

Deep learning models for plant disease recognition often perform well on curated laboratory datasets but degrade substantially in real-world field conditions due to domain shift. This paper presents a reproducible lab-to-field evaluation of cross-domain transfer from PlantVillage to PlantDoc, using a conservative eight-class alignment protocol and a curated clean target subset. We compare random balanced sampling with diversity-based sampling under a fixed training budget and evaluate multiple backbones, including ResNet18, EfficientNet-B0, and ViT-B/16. Results show that diversity-aware sampling improves transfer robustness, raising accuracy from 21.4% to 26.8% and macro-F1 from 17.3% to 22.1%, while larger models do not consistently improve generalization. The findings highlight that data-centric strategies, especially careful curation and diversity-based selection, are more effective than increasing model capacity alone for robust field deployment of plant disease recognition systems.

Deep learning models for plant disease recognition often perform well on curated laboratory datasets but degrade substantially in real-world field conditions due to domain shift. This paper presents a reproducible lab-to-field evaluation of cross-domain transfer from PlantVillage to PlantDoc, using a conservative eight-class alignment protocol and a curated clean target subset. We compare random balanced sampling with diversity-based sampling under a fixed training budget and evaluate multiple backbones, including ResNet18, EfficientNet-B0, and ViT-B/16. Results show that diversity-aware sampling improves transfer robustness, raising accuracy from 21.4% to 26.8% and macro-F1 from 17.3% to 22.1%, while larger models do not consistently improve generalization. The findings highlight that data-centric strategies, especially careful curation and diversity-based selection, are more effective than increasing model capacity alone for robust field deployment of plant disease recognition systems.

Deep learning models for plant disease recognition often rely on curated laboratory datasets that fail to capture the variability and noise present in real agricultural environments. This paper proposes a symptom-aware protocol for collecting and annotating plant disease datasets designed for field-ready deployment. The protocol emphasizes realistic acquisition conditions, symptom diversity, multi-stage annotation, and quality control procedures to reduce label noise and dataset bias. We describe the complete pipeline, including image acquisition guidelines, symptom-driven categorization, annotation validation, and dataset organization for machine learning workflows. Experimental observations highlight how careful dataset design improves robustness and transferability of plant disease recognition models when moving from controlled datasets to real-world field scenarios. The proposed protocol aims to support the creation of reliable, scalable datasets for agricultural computer vision and to facilitate reproducible evaluation of plant disease detection systems.

Deep learning models for plant disease recognition often rely on curated laboratory datasets that fail to capture the variability and noise present in real agricultural environments. This paper proposes a symptom-aware protocol for collecting and annotating plant disease datasets designed for field-ready deployment. The protocol emphasizes realistic acquisition conditions, symptom diversity, multi-stage annotation, and quality control procedures to reduce label noise and dataset bias. We describe the complete pipeline, including image acquisition guidelines, symptom-driven categorization, annotation validation, and dataset organization for machine learning workflows. Experimental observations highlight how careful dataset design improves robustness and transferability of plant disease recognition models when moving from controlled datasets to real-world field scenarios. The proposed protocol aims to support the creation of reliable, scalable datasets for agricultural computer vision and to facilitate reproducible evaluation of plant disease detection systems.

Control charts do more than raise an alarm: their shapes can give an early indication of what has changed in a process. This study considers the case in which one chart window contains more than one abnormal behavior. The observed sequence is then a mixture rather than a pure pattern. We formulate this problem directly as multi-label classification. A one-dimensional CNN receives a raw-scale window of 32 observations and predicts the active elementary labels. The controlled protocol contains twelve scenarios: normal behavior, six single abnormal patterns, and five selected concurrent patterns. Raw-scale input is retained because shift patterns depend partly on level information that may be weakened by window-wise normalization. The retained training setup gives additional exposure to difficult shift and trend cases, while validation and testing remain balanced. Across five repeated trainings, the model achieved 96.11% exact match accuracy, 96.41% precision, 96.46% recall, 96.44% F1-score, and 1.04% Hamming loss. The 95% confidence interval for exact match was 96.05–96.17%. Additional analyses show stable performance around a decision threshold of 0.5, strong cyclic and systematic recognition, and lower performance for short shift cases. The results support direct multi-label CNN recognition for the selected protocol. Broader shift-containing mixtures, more complex combinations, varying noise conditions, and real industrial validation remain outside the scope of the present controlled study.

This record provides supplementary research artifacts supporting the study "When Reliability Changes the Attack Surface: Decision-Oracle Geometry of Error-Corrected Match-on-Card Biometrics." The archive contains derived experimental result artifacts and Tamarin proof outputs supporting the decoder-conformance, coding-regime, decision-oracle extraction, helper-composition, cold-start sensitivity, and formal capture-to-probe admission analyses reported in the manuscript. The experimental artifacts include operating-point results, decoder-conformance measurements, coding-regime transition data, primary oracle-extraction results, confidence intervals, helper-parity analyses, oracle/helper composition results, and cold-start sensitivity results. The formal-verification artifacts contain proof outputs for the authenticated-channel-only negative control, generic-signer negative control, restricted capture-admission model, and capture-key-compromise model, together with run-status, tool-version, and manifest information. The MORPH source images are subject to licensing restrictions and are not redistributed. This record contains no MORPH source images or subject-level source imagery. The archive is intended to support inspection and reproducibility of the empirical and formal results reported in the associated manuscript.

In the era of pervasive cyber threats and exponential growth in digital services, the inadequacy of single-factor authentication has become increasingly evident. Multi-Factor Authentication (MFA), which combines knowledge-based factors (passwords, PINs), possession-based factors (smart cards, tokens), and inherence-based factors (biometric traits), has emerged as a robust defense mechanism. Recent breakthroughs in deep learning have transformed the capabilities of biometric systems, enabling higher accuracy, resilience to spoofing, and seamless integration with hardware-based solutions. At the same time, smart card technologies have evolved to include on-chip biometric verification, cryptographic processing, and secure storage, thereby enabling compact and secure multi-factor devices. This survey presents a comprehensive synthesis of recent work (2019-2025) at the intersection of deep learning, biometrics, and smart card technologies for MFA. We analyze biometric modalities (face, fingerprint, iris, voice), review hardware-based approaches (smart cards, NFC, TPMs, secure enclaves), and highlight integration strategies for real-world applications such as digital banking, healthcare IoT, and critical infrastructure. Furthermore, we discuss the major challenges that remain open, including usability-security tradeoffs, adversarial attacks on deep learning models, privacy concerns surrounding biometric data, and the need for standardization in MFA deployment. By consolidating current advancements, limitations, and research opportunities, this survey provides a roadmap for designing secure, scalable, and user-friendly authentication frameworks.

We present a practical match-on-card design for face verification in which compact 64/128-bit templates are produced off-card by PCA-ITQ and compared on-card via constant-time Hamming distance. We specify ISO/IEC 7816-4 and 14443-4 command APDUs with fixed-length payloads and decision-only status words (no score leakage), together with a minimal per-identity EEPROM map. Using real binary codes from a CelebA working set (55 identities, 412 images), we (i) derive operating thresholds from ROC/DET, (ii) replay enroll->verify transactions at those thresholds, and (iii) bound end-to-end time by pure link latency plus a small constant on-card budget. Even at the slowest contact rate (9.6 kbps), total verification time is 43.9 ms (64 b) and 52.3 ms (128 b); at 38.4 kbps both are <14 ms. At FAR = 1%, both code lengths reach TPR = 0.836, while 128 b lowers EER relative to 64 b. An optional +6 B helper (targeted symbol-level parity over empirically unstable bits) is latency-negligible. Overall, short binary templates, fixed-payload decision-only APDUs, and constant-time matching satisfy ISO/IEC transport constraints with wide timing margin and align with ISO/IEC 24745 privacy goals. Limitations: single-dataset evaluation and design-level (pre-hardware) timing; we outline AgeDB/CFP-FP and on-card microbenchmarks as next steps.

Compact 64 bit binary face templates derived from deep features enable constant-time Hamming matching and tiny on-card storage for contact/contactless smart cards. In practice, verification errors at low FAR are dominated by a small set of systematically unstable bit positions. We propose targeted parity, which applies a short Reed-Solomon (RS) code only to the 𝑊 least-stable bits while leaving the remaining bits unchanged for discriminability. With 𝑊=16, 𝑚=2-bit symbols, and 𝑝 ∈ {4, 6, 8} parity symbols, the helper data per identity is just 𝑝 bytes (no per-user bit-index storage). On 64 bit templates obtained via ArcFace→PCA→ITQ, targeted RS improves robustness consistently across enrollment sizes; the best configuration attains TPR@FAR=1% = 0.942 at 𝑛=5 images with 8 B helper and negligible decoding latency. Compared to masking the 𝐾 most-stable bits (same logical template length), targeted RS yields higher TPR while preserving discriminative capacity. We provide the algorithm, storage/latency analysis, and ablations over 𝑊, symbol size 𝑚, parity 𝑝, and enrollment 𝑛.

Facial recognition systems are increasingly deployed in privacy-sensitive and resource-constrained environments such as smart cards. However, traditional face verification relies on high-dimensional floating-point embeddings, which are unsuitable for compact and efficient matching on such platforms. To address this challenge, this work investigates the generation of binary face templates that retain identity information while reducing storage and computational cost. The objective of this study is to benchmark binary biometric representations derived from mobile-class convolutional neural networks (CNNs), aiming to support reproducible, lightweight face verification pipelines. We evaluate four lightweight CNNs—EfficientNet-B0, MobileNetV2, ShuffleNetV2, and SqueezeNet1_1—trained on the MORPH dataset. Binary templates are generated via Principal Component Analysis followed by Iterative Quantization (PCA–ITQ) at 32, 64, and 128 bits. Models are tested cross-dataset on the Georgia Tech Face Database (GT Face) to assess generalization. At 128 bits, EfficientNet-B0 and MobileNetV2 achieve strong verification performance, with area under the curve (AUC) ≈ 0.895–0.899 and equal error rate (EER) ≈ 0.182–0.185. A Hamming-distance analysis confirms clear separation between genuine and impostor pairs, and the bit-flip rate (~17%) indicates intra-subject consistency. Bit-length scaling further reveals monotonic improvements in AUC from 32 to 128 bits, highlighting a trade-off between accuracy and compactness. These results demonstrate that binary templates from lightweight CNNs can deliver efficient, privacy-preserving authentication with limited performance degradation. The proposed pipeline supports reproducibility and aligns with FAIR data principles, making it suitable for secure biometric deployments on constrained hardware.

We present a practical match-on-card design for face verification in which compact 64/128-bit templates are produced off-card by PCA-ITQ and compared on-card via constant-time Hamming distance. We specify ISO/IEC 7816-4 and 14443-4 command APDUs with fixed-length payloads and decision-only status words (no score leakage), together with a minimal per-identity EEPROM map. Using real binary codes from a CelebA working set (55 identities, 412 images), we (i) derive operating thresholds from ROC/DET, (ii) replay enroll->verify transactions at those thresholds, and (iii) bound end-to-end time by pure link latency plus a small constant on-card budget. Even at the slowest contact rate (9.6 kbps), total verification time is 43.9 ms (64 b) and 52.3 ms (128 b); at 38.4 kbps both are <14 ms. At FAR = 1%, both code lengths reach TPR = 0.836, while 128 b lowers EER relative to 64 b. An optional +6 B helper (targeted symbol-level parity over empirically unstable bits) is latency-negligible. Overall, short binary templates, fixed-payload decision-only APDUs, and constant-time matching satisfy ISO/IEC transport constraints with wide timing margin and align with ISO/IEC 24745 privacy goals. Limitations: single-dataset evaluation and design-level (pre-hardware) timing; we outline AgeDB/CFP-FP and on-card microbenchmarks as next steps.

Compact 64 bit binary face templates derived from deep features enable constant-time Hamming matching and tiny on-card storage for contact/contactless smart cards. In practice, verification errors at low FAR are dominated by a small set of systematically unstable bit positions. We propose targeted parity, which applies a short Reed-Solomon (RS) code only to the 𝑊 least-stable bits while leaving the remaining bits unchanged for discriminability. With 𝑊=16, 𝑚=2-bit symbols, and 𝑝 ∈ {4, 6, 8} parity symbols, the helper data per identity is just 𝑝 bytes (no per-user bit-index storage). On 64 bit templates obtained via ArcFace→PCA→ITQ, targeted RS improves robustness consistently across enrollment sizes; the best configuration attains TPR@FAR=1% = 0.942 at 𝑛=5 images with 8 B helper and negligible decoding latency. Compared to masking the 𝐾 most-stable bits (same logical template length), targeted RS yields higher TPR while preserving discriminative capacity. We provide the algorithm, storage/latency analysis, and ablations over 𝑊, symbol size 𝑚, parity 𝑝, and enrollment 𝑛.

In the era of pervasive cyber threats and exponential growth in digital services, the inadequacy of single-factor authentication has become increasingly evident. Multi-Factor Authentication (MFA), which combines knowledge-based factors (passwords, PINs), possession-based factors (smart cards, tokens), and inherence-based factors (biometric traits), has emerged as a robust defense mechanism. Recent breakthroughs in deep learning have transformed the capabilities of biometric systems, enabling higher accuracy, resilience to spoofing, and seamless integration with hardware-based solutions. At the same time, smart card technologies have evolved to include on-chip biometric verification, cryptographic processing, and secure storage, thereby enabling compact and secure multi-factor devices. This survey presents a comprehensive synthesis of recent work (2019-2025) at the intersection of deep learning, biometrics, and smart card technologies for MFA. We analyze biometric modalities (face, fingerprint, iris, voice), review hardware-based approaches (smart cards, NFC, TPMs, secure enclaves), and highlight integration strategies for real-world applications such as digital banking, healthcare IoT, and critical infrastructure. Furthermore, we discuss the major challenges that remain open, including usability-security tradeoffs, adversarial attacks on deep learning models, privacy concerns surrounding biometric data, and the need for standardization in MFA deployment. By consolidating current advancements, limitations, and research opportunities, this survey provides a roadmap for designing secure, scalable, and user-friendly authentication frameworks.

Facial recognition systems are increasingly deployed in privacy-sensitive and resource-constrained environments such as smart cards. However, traditional face verification relies on high-dimensional floating-point embeddings, which are unsuitable for compact and efficient matching on such platforms. To address this challenge, this work investigates the generation of binary face templates that retain identity information while reducing storage and computational cost. The objective of this study is to benchmark binary biometric representations derived from mobile-class convolutional neural networks (CNNs), aiming to support reproducible, lightweight face verification pipelines. We evaluate four lightweight CNNs—EfficientNet-B0, MobileNetV2, ShuffleNetV2, and SqueezeNet1_1—trained on the MORPH dataset. Binary templates are generated via Principal Component Analysis followed by Iterative Quantization (PCA–ITQ) at 32, 64, and 128 bits. Models are tested cross-dataset on the Georgia Tech Face Database (GT Face) to assess generalization. At 128 bits, EfficientNet-B0 and MobileNetV2 achieve strong verification performance, with area under the curve (AUC) ≈ 0.895–0.899 and equal error rate (EER) ≈ 0.182–0.185. A Hamming-distance analysis confirms clear separation between genuine and impostor pairs, and the bit-flip rate (~17%) indicates intra-subject consistency. Bit-length scaling further reveals monotonic improvements in AUC from 32 to 128 bits, highlighting a trade-off between accuracy and compactness. These results demonstrate that binary templates from lightweight CNNs can deliver efficient, privacy-preserving authentication with limited performance degradation. The proposed pipeline supports reproducibility and aligns with FAIR data principles, making it suitable for secure biometric deployments on constrained hardware.