An Industrial-Grade Solution for Crop Disease Image Detection Tasks

Abstract

Crop leaf diseases can reflect the current health status of the crop, and the rapid and automatic detection of field diseases has become one of the difficulties in the process of industrialization of agriculture. In the widespread application of various machine learning techniques, recognition time consumption and accuracy remain the main challenges in moving agriculture toward industrialization. This article proposes a novel network architecture called YOLO V5-CAcT to identify crop diseases. The fast and efficient lightweight YOLO V5 is chosen as the base network. Repeated Augmentation, FocalLoss, and SmoothBCE strategies improve the model robustness and combat the positive and negative sample ratio imbalance problem. Early Stopping is used to improve the convergence of the model. We use two technical routes of model pruning, knowledge distillation and memory activation parameter compression ActNN for model training and identification under different hardware conditions. Finally, we use simplified operators with INT8 quantization for further optimization and deployment in the deep learning inference platform NCNN to form an industrial-grade solution. In addition, some samples from the Plant Village and AI Challenger datasets were applied to build our dataset. The average recognition accuracy of 94.24% was achieved in images of 59 crop disease categories for 10 crop species, with an average inference time of 1.563 ms per sample and model size of only 2 MB, reducing the model size by 88% and the inference time by 72% compared with the original model, with significant performance advantages. Therefore, this study can provide a solid theoretical basis for solving the common problems in current agricultural disease image detection. At the same time, the advantages in terms of accuracy and computational cost can meet the needs of agricultural industrialization.

Keywords: activate quantitative; convolutional neural network; crop disease detection; knowledge distillation; model compression; model deployment.

Figure 1

Technical route diagram.

Figure 2

Crop disease YOLOV5 network structure diagram.

Figure 3

FPGM pruning process.

Figure 4

Step of knowledge distillation.

Figure 5

ActNN compression and decompression process.

Figure 6

Integration of YOLO V5 model structure using ActNN.

Figure 7

layer fusion and data reuse.

Figure 8

Sample images of the PFD dataset.

Figure 9

Confidence accuracy curve and recall accuracy curve. (A) Precision-confidence curves of the YOLO V5s-CAcT3 model on the PFD dataset. (B) PR curves of YOLO V5s-CAcT3 model on the PFD dataset.

Figure 10

(A) AI Challenger accuracy, parameter drop, and sparsity curves vs. (B) PFD accuracy, parameter drop, and sparsity curves.

Figure 11

Histogram of model weight change during sparse training.

Figure 12

Effect of temperature parameters on knowledge distillation.

Figure A1

Confusion matrix for the predictions of the best model trained on the PFD dataset.

Figure A2

Prediction of the best model trained on the PFD dataset and live labeling of the sample images.

Figure A3

Prediction of the best model trained on the PFD dataset and live labeling of the sample images.