Table 1 The following table explores the study in terms of author and year of publication (reference), objectives of the study, dataset used for the study, results of the study and limitations of the study.
References | Objectives | Dataset used | Results | Limitations |
|---|---|---|---|---|
Elaraby et al., 2022b20 | DL model for stress detection in five different crops, i.e., cucumber, corn, wheat, grape, cotton | 54k images of fourteen different crops from Plant Village dataset | Accuracy 98.83%, Sensitivity (sens.) 98.78%, F Score 98.47%, Precision (precision) 98.67%, and Specificity (spec.) 98.53% | Real-time implementation testing required |
Pan et al., 202421 | CDDLite-YOLO model for cotton plant stress detection + enhancing acc. with minimum YOLO parameters | 1530 natural field images of cotton plant (38% verticillium wilt, 34% anthracnose, and 28% fusarium wilt) | (Mean Average Precision) mAP achieved is 90.60% with 3.6G FLOPS and 1.8 M parameters + identification speed of 222.22 FPS | Only 78.10% mAP for verticillium detection |
Ahmed, 202122 | Transfer learning based Custom CNN i.e., DCPLD-CNN for cotton leaf disease detection. | The dataset used in the Cotton Plant and Leaf Disease recognition study was collected from a specified source [38] | Accuracy 98.77% + validation acc. of 88.99% and 98.77% for 100 and 500 iterations, respectively | More robust testing specific to cotton plant dataset needed |
Gao et al., 202423 | DL model for cotton plant stress detection | Datasets for the study were collected from diversified sources through manual collection and internet crawling techniques | 94% accuracy, 95% mAP, and speed of 49.7 FPS | Further study is needed to maintain performance on larger datasets and reduce computational resource consumption |
Bharathi et al., 202424 | Random tree based- adaptive fire-hawk DL model i.e., DQRR-AFH | Hybrid database by combining 1710 natural field and internet images of cotton plant | 98.88% accuracy, 99.21% F1 score, 97% precision + Performance comparison with WL-CNN, ECPRC, and DT models | Only two class classifications are performed |
Li et al., 202425 | CFNet-VoV-GCSP-LSKNet-YOLOv8s model for cotton stress detection | 6 public datasets from Kaggel | 89.9% precision, 90.70% recall, and 93.7% mAP(0.5) | The study does not address the limitations explicitly, focusing more on the proposed method’s enhancements and performance metrics. |
Nazeer et al., 202426 | Develop a dataset of cotton leaf images to support automated disease detection systems + DL model for detection of Cotton Leaf Curl Disease (CLCuD) | Hybrid dataset of Kaggle images and natural self-collected 1349 images of cotton leaf | 99% accuracy achieved | Model restricted to Leaf curl disease in cotton plant |
Latif et al., 20218 | Develop an automated technique for detecting cotton leaf diseases using DL | The study utilized 1000 self-collected datasets of cotton diseases labelled and augmented by an expert for training and testing purposes | Achieved an accuracy of 98.8% using Cubic SVM | The model applied to four classes only, which are Areolate Mildew, Myrothecium leaf spot, and Soreshine |
Kolachi et al., 202327 | Identification of blight and curl disease in cotton plant using a custom YOLO DL model | Natural dataset from farmer’s fields of Sindh, Pakistan. It consists of healthy leaves, bacterial blight, and curl virus images. 1000 images were sourced from Kaggle, GitHub, and Google to enhance the dataset’s diversity and size. After augmentation, a total of 5046 images were formed | The YOLOv5 model achieved 92% accuracy in disease classification | Only two class classifications are performed: bacterial blight and curl virus |
Zhu et al., 202228 | Develop a cotton disease identification method based on pruning on VGG16, ResNet164, and DenseNet40 to address deployability issues on resource-limited smart devices | PlantVillage consists of 14 types of plants with 54,306 images of healthy and diseased leaves | The compressed models (size 2.2 MB) achieved high accuracies, with DenseNet40-80-T achieving 97.23% | Lack of detailed discussion on the potential challenges or drawbacks of the proposed pruning algorithm and compression strategies used for cotton disease identification based on DCNN models |
Thivya et al., 20249 | Develop a novel DL pipeline, CoDet, for cotton plant detection and disease identification | 30k images collected from internet (25k for training and rest 5k for validation testing) | CoDet outperformed other models in comparative study using different matrics | More robust testing needed considering Indian environment and physiochemical traits of cotton plant |
Rai and Pahuja, 202329 | Deep-CNN for cotton stress detection | Hybrid dataset of 2293 natural images and Kaggle images | 97.98% accuracy achieved | Real time performance testing needed |
Rai and Pahuja, 202430 | Using DL methods for cotton stress identification | Two different datasets from Kaggle of 2310 and 1711 images respectively | 99.48% accuracy and 99% sens. achieved | More robust testing needed |
Shahid et al., 202431 | GoogleNet, VGG19, AlexNet, and InceptionV3 for identification of cotton plant stress | Natural dataset from Balochistan, Pakistan covering all the 3 phases i.e., sowing, germination and maturity | Accuracy achieved by GoogleNet, AlexNet, and InceptionV3 is 93.40%, 93.40%, and 91.80% respectively | The accuracy is low as compared to the other models |
Kukadiya et al., 202410 | Pre-trained VGG16 and InceptionV3 used to detect early cotton leaf diseases | 1786 images of 4 different types i.e., blight, curl, wilt and healthy from PlantVillage dataset | Training and Testing accuracy of 98% and 95% achieved respectively | It focused only on four cotton diseases |
Islam et al., 202319 | Hybrid DL models by combining Transfer learning with Xception, Inception V3, VGG 16 and 19 | 2310 images from Kaggle dataset | VGG-16 achieved an accuracy of 90.22%, while VGG-19, Inception-V3, and Xception achieved higher accuracies of 96.74%, 97.83%, and 98.70%, respectively | Only binary classification achieved, more robust testing needed |
