Deep Learning Models Compression for Agricultural Plants

Abstract

Deep learning has been successfully showing promising results in plant disease detection, fruit counting, yield estimation, and gaining an increasing interest in agriculture. Deep learning models are generally based on several millions of parameters that generate exceptionally large weight matrices. The latter requires large memory and computational power for training, testing, and deploying. Unfortunately, these requirements make it difficult to deploy on low-cost devices with limited resources that are present at the fieldwork. In addition, the lack or the bad quality of connectivity in farms does not allow remote computation. An approach that has been used to save memory and speed up the processing is to compress the models. In this work, we tackle the challenges related to the resource limitation by compressing some state-of-the-art models very often used in image classification. For this we apply model pruning and quantization to LeNet5, VGG16, and AlexNet. Original and compressed models were applied to the benchmark of plant seedling classification (V2 Plant Seedlings Dataset) and Flavia database. Results reveal that it is possible to compress the size of these models by a factor of 38 and to reduce the FLOPs of VGG16 by a factor of 99 without considerable loss of accuracy.

Classification of Plant Leaves Using New Compact Convolutional Neural Network Models

“…For the VGG16 and VGG 19 models, the accuracy achieved by of models is 81.3% and 96.25%, respectively [ 49 ]. The combination of pruning and post-quantization was applied to VGG16, AlexNet, and LeNet model [ 50 ]. The pruning step was responsible for reducing the model size.…”

Fruit quality and defect image classification with conditional GAN data augmentation

“…Considering that processing is performed on large quantities of fruit and that there may be time and energy restrictions, model pruning is performed on the network to explore the possibility of smaller model sizes, more apt for real-world usage [49,50], through polynomial decay. For each model, 9 pruning experiments were performed with weight sparsity ranging from 0.9 (10% of original size) to 0.1 (90% of original size).…”

A Deep Learning Model Compression and Ensemble Approach for Weed Detection

“…As demonstrated in Figure 1 from the work of Han et al [29], all three compression techniques under the right conditions retain the prediction accuracy of the original model. Regardless, some studies have found that the pruning ratio affects the accuracy of the model [38], [39]. In effect, a slight reduction in accuracy is possible depending on the percentage of the model's trainable weights that are pruned [38]- [40].…”

“…Regardless, some studies have found that the pruning ratio affects the accuracy of the model [38], [39]. In effect, a slight reduction in accuracy is possible depending on the percentage of the model's trainable weights that are pruned [38]- [40]. The current study makes the following contributions: theoretically, the study presents an approach that combines well-known DL techniques to reduce model complexity such that they require less expensive equipment to run without the performance degradation demonstrated in past studies; and practically, the proposed approach which solves some of the issues with DL at the edge could be employed in different contexts other than precision agriculture.…”