I need improvements on Python code for CNN $($convolutional neural network$)$model that can follow the steps:

$1.$Recognize the crop;

$2.$Predict healthy$/$unhealthy crop;

$3.$Detect the disease;

$4.$If the crop is not recognized, try to predict healthy$/$unhealthy and then detect the disease.

Data from kaggle.com: $14$kinds of crop, $43459$images for training and $10876$images for testing.

Print an accuracy plot diagram for each model. Also, logic needs to be corrected and final result printed. Please, leave comments for all steps.

import tensorflow as tf from tensorflow.keras.preprocessing.image import ImageDataGenerator from tensorflow.keras.models import Sequential from tensorflow.keras.layers import Conv$2$D$,$ MaxPooling$2$D$,$ Flatten, Dense, Dropout from tensorflow.keras.callbacks import EarlyStopping from tensorflow.keras.preprocessing import image import matplotlib.pyplot as plt import numpy as np import os train$\_$dir $=\text{'}/$train$\text{'}$ test$\_$dir $=\text{'}/$test$\text{'}$ IMG$\_$HEIGHT $=150$ IMG$\_$WIDTH $=150$ BATCH$\_$SIZE $=32$ EPOCHS $=10$ train$\_$datagen $=$ ImageDataGenerator$($ rescale $=1\mathrm{.}0/255,$ rotation$\_$range $=20,$ width$\_$shift$\_$range $=0\mathrm{.}2,$ height$\_$shift$\_$range $=0\mathrm{.}2,$ shear$\_$range $=0\mathrm{.}2,$ zoom$\_$range $=0\mathrm{.}2,$ horizontal$\_$flip $=$ True, fill$\_$mode $=$ 'nearest' $)$ test$\_$datagen $=$ ImageDataGenerator$($rescale$=1\mathrm{.}0/255)$ train$\_$data $=$ train$\_$datagen.flow$\_$from$\_$directory$($ train$\_$dir, target$\_$size $=($IMG$\_$HEIGHT, IMG$\_$WIDTH$),$ batch$\_$size $=$ BATCH$\_$SIZE, class$\_$mode $=$ 'categorical' $)$ test$\_$data $=$ test$\_$datagen.flow$\_$from$\_$directory$($ test$\_$dir, target$\_$size $=($IMG$\_$HEIGHT, IMG$\_$WIDTH$),$ batch$\_$size $=$ BATCH$\_$SIZE, class$\_$mode $=$ 'categorical' $)$ def create$\_$model$($output$\_$classes$)$: model $=$ Sequential$([$ Conv$2$D$(32,(3,3),$ activation $=$ 'relu', input$\_$shape $=($IMG$\_$HEIGHT, IMG$\_$WIDTH, $3)),$ MaxPooling$2$D$(2,2),$ Conv$2$D$(64,(3,3),$ activation $=$ 'relu'$),$ MaxPooling$2$D$(2,2),$ Conv$2$D$(128,(3,3),$ activation $=$ 'relu'$),$ MaxPooling$2$D$(2,2),$ Conv$2$D$(128,(3,3),$ activation $=$ 'relu'$),$ MaxPooling$2$D$(2,2),$ Flatten$(),$ Dense$(512,$ activation $=$ 'relu'$),$ Dropout$(0\mathrm{.}5),$ Dense$($output$\_$classes, activation $=$ 'softmax'$)])$ return model crop$\_$model $=$ create$\_$model$($output$\_$classes $=14)$ crop$\_$model.compile$($optimizer $=$ 'adam', loss $=$ 'categorical$\_$crossentropy', metrics $=[\text{'}$accuracy$\text{'}])$ crop $=$ crop$\_$model.fit$($test$\_$data, validation$\_$data $=$ train$\_$data, epochs $=$ EPOCHS$)$ plt$.$plot$($crop$.$history$[\text{'}$accuracy$\text{'}],$ label $=$ 'Train Accuracy'$)$ plt$.$plot$($crop$.$history$[\text{'}$val$\_$accuracy'$],$ label $=$ 'Validation Accuracy'$)$ plt$.$xlabel$(\text{'}$Epochs$\text{'})$ plt$.$ylabel$(\text{'}$Accuracy$\text{'})$ plt$.$legend$()$ plt$.$show$()$ test$\_$loss, test$\_$accuracy $=$ crop$\_$model.evaluate$($test$\_$data$)$ print$($f$\text{'}$Test Accuracy: $\{$test$\_$accuracy $*100$:$\mathrm{.}2$f$\}\%\text{'})$ health$\_$model $=$ create$\_$model$($output$\_$classes $=2)$ #Healthy$/$Unhealthy health$\_$model.compile$($optimizer $=$ 'adam', loss $=$ 'categorical$\_$crossentropy', metrics $=[\text{'}$accuracy$\text{'}])$ health $=$ health$\_$model.fit$($test$\_$data, validation$\_$data $=$ train$\_$data, epochs $=$ EPOCHS$)$ plt$.$plot$($health$.$history$[\text{'}$accuracy$\text{'}],$ label $=$ 'Train Accuracy'$)$ plt$.$plot$($health$.$history$[\text{'}$val$\_$accuracy'$],$ label $=$ 'Validation Accuracy'$)$ plt$.$xlabel$(\text{'}$Epochs$\text{'})$ plt$.$ylabel$(\text{'}$Accuracy$\text{'})$ plt$.$legend$()$ plt$.$show$()$ disease$\_$model $=$ create$\_$model$($output$\_$classes $=5)$ #$5$ diseases disease$\_$model.compile$($optimizer $=$ 'adam', loss $=$ 'categorical$\_$crossentropy', metrics $=[\text{'}$accuracy$\text{'}])$ disease $=$ disease$\_$model.fit$($test$\_$data, validation$\_$data $=$ train$\_$data, epochs $=$ EPOCHS$)$ plt$.$plot$($disease$.$history$[\text{'}$accuracy$\text{'}],$ label $=$ 'Train Accuracy'$)$ plt$.$plot$($disease$.$history$[\text{'}$val$\_$accuracy'$],$ label $=$ 'Validation Accuracy'$)$ plt$.$xlabel$(\text{'}$Epochs$\text{'})$ plt$.$ylabel$(\text{'}$Accuracy$\text{'})$ plt$.$legend$()$ plt$.$show$()$ def classify$\_$image$($image$\_$path$)$: image$\_$path $=\text{'}?\text{'}$ img $=$ image.load$\_$img$($image$\_$path, target$\_$size$=($IMG$\_$HEIGHT, IMG$\_$WIDTH$))$ img$\_$array $=$ image.img$\_$to$\_$array$($img$)/255\mathrm{.}0$ img$\_$array $=$ np$.$expand$\_$dims$($img$\_$array, axis$=0)$ crop$\_$prediction $=$ crop$\_$model.predict$($img$\_$array$)$ crop$\_$label $=$ np$.$argmax$($crop$\_$prediction$)$ if crop$\_$label $==-1$: health$\_$prediction $=$ health$\_$model.predict$($img$\_$array$)$ health$\_$label $=$ np$.$argmax$($health$\_$prediction$)$ if health$\_$label $==1$: disease$\_$prediction $=$ disease$\_$model.predict$($img$\_$array$)$ disease$\_$label $=$ np$.$argmax$($disease$\_$prediction$)$ return f$"$Unknown crop, Unhealthy, Disease detected: $\{$disease$\_$label$\}"$ else: return "Unknown crop, Healthy" else: health$\_$prediction $=$ health$\_$model.predict$($img$\_$array$)$ health$\_$label $=$ np$.$argmax$($health$\_$prediction$)$ if health$\_$label $==1$: disease$\_$prediction $=$ disease$\_$model.predict$($img$\_$array$)$ disease$\_$label $=$ np$.$argmax$($disease$\_$prediction$)$ return f$"$Crop: $\{$crop$\_$label$\},$ Unhealthy, Disease detected: $\{$disease$\_$label$\}"$ else: return f$"$Crop: $\{$crop$\_$label$\},$ Healthy" result $=$ classify$\_$image$(\text{'}?\text{'})$ print$($result$)$ crop$\_$model.save$(\text{'}$crop$\_$recognition$\_$model.h$5\text{'})$ health$\_$model.save$(\text{'}$health$\_$recognition$\_$model.h$5\text{'})$ disease$\_$model.save$(\text{'}$disease$\_$recognition$\_$model.h$5\text{'})$